Skip to content

gemv: add optional fused GELU epilogue (NPU2)#135

Open
atassis wants to merge 1 commit into
amd:develfrom
atassis:feat/gemv-gelu-epilogue
Open

gemv: add optional fused GELU epilogue (NPU2)#135
atassis wants to merge 1 commit into
amd:develfrom
atassis:feat/gemv-gelu-epilogue

Conversation

@atassis

@atassis atassis commented Jul 10, 2026

Copy link
Copy Markdown
Contributor

Adds an opt-in epilogue parameter to GEMV. epilogue="none" (default) is byte-identical to today; epilogue="gelu" fuses a GELU (tanh approximation) over each output tile inside the producing core, folding an activation that would otherwise be a separate elementwise pass over the GEMV output.

Details

  • The GELU runs once per full m_output tile in core_body (after the matvec inner-loop has filled all rows), not per matvec call whose m_input tile can be smaller than the 16-wide activation vector.
  • No duplicated math. aie_kernels/aie2p/gelu.cc is refactored so the per-16-lane computation is a shared helper used by both the existing out-of-place gelu_tanh_approx_bf16 and a new in-place gelu_tile_bf16. In-place is aliasing-correct (a single restrict pointer, each slot read then written); reusing the out-of-place kernel with input == output would violate its restrict contract, so a dedicated in-place entry point is the correct way to share the code.
  • Since gelu.cc lives in aie2p/, the fused epilogue is NPU2-only; GEMV raises NotImplementedError if it is requested on NPU1. When enabled, the core links a (matvec, gelu) archive; the default path still links the single matvec object.

Test

test_gemv_gelu compares the fused output against a gelu(A @ B) golden across three shapes, using the same latency/bandwidth/throughput @pytest.mark.metrics as test_gemv. Verified on NPU2: 15/15 pass.

This is opt-in and default-preserving; it also composes with the existing func_prefix fusion support (#123) for use as a fused block.

@andrej andrej left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is a good addition and a natural fusion opportunity. If you have any interest in this, I think it would be cool to think about if we can generalize this across other activation types as well -- sort of a generic "run this kernel in-place on the results" flag, where maybe you could pass an arbitrary Kernel object. That would be a follow-on PR of course.

@github-actions

github-actions Bot commented Jul 13, 2026

Copy link
Copy Markdown
Contributor

CI Test Results

fb80db1 (2026_07_13_22_30_12)

IRON - CI Summary

Small

iron/operators/axpy
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_axpy[input_length_2048-num_aie_columns_1-tile_size_2048-scalar_factor_3.0] 139.10 303.20
test_axpy[input_length_2048-num_aie_columns_2-tile_size_1024-scalar_factor_3.0] 188.50 257.80
test_axpy[input_length_2048-num_aie_columns_4-tile_size_512-scalar_factor_3.0] 168.08 336.72
test_axpy[input_length_2048-num_aie_columns_8-tile_size_256-scalar_factor_3.0] 183.62 - -
iron/operators/dequant
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_dequant[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-group_size_32] 178.70 354.66
test_dequant[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-group_size_32] 154.00 446.40
test_dequant[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-group_size_32] 177.76 358.38
test_dequant[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-group_size_32] 173.76 540.84
test_dequant[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-group_size_32] 179.14 389.54
test_dequant[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-group_size_32] 229.88 494.50
test_dequant[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-group_size_32] 250.58 - -
test_dequant[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128-group_size_32] 238.26 - -
iron/operators/elementwise_add
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_elementwise_add[input_length_2048-num_aie_columns_1-tile_size_2048] 175.36 315.96
test_elementwise_add[input_length_2048-num_aie_columns_2-tile_size_1024] 182.20 364.80
test_elementwise_add[input_length_2048-num_aie_columns_4-tile_size_512] 194.68 314.92
test_elementwise_add[input_length_2048-num_aie_columns_8-tile_size_256] 192.96 - -
iron/operators/elementwise_mul
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_elementwise_mul[input_length_2048-num_aie_columns_1-tile_size_2048] 206.50 283.04
test_elementwise_mul[input_length_2048-num_aie_columns_2-tile_size_1024] 157.22 743.46
test_elementwise_mul[input_length_2048-num_aie_columns_4-tile_size_512] 181.88 419.90
test_elementwise_mul[input_length_2048-num_aie_columns_8-tile_size_256] 207.82 - -
iron/operators/gelu
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_gelu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] 162.40 342.16
test_gelu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024] 183.58 370.56
test_gelu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] 170.90 430.00
test_gelu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512] 183.94 962.78
test_gelu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] 177.60 313.56
test_gelu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256] 185.08 397.18
test_gelu[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256] 200.14 - -
test_gelu[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128] 206.60 - -
iron/operators/gemm
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_gemm[M_1792-K_896-N_1152-num_aie_columns_8-b_col_maj_False-c_col_maj_True-m_64-k_32-n_48-trace_size_0-partition_N_1] 2319.82 - -
test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_False-c_col_maj_False-m_48-k_96-n_16-trace_size_0-partition_N_1] 254.58 591.86
test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_True-c_col_maj_True-m_48-k_96-n_16-trace_size_0-partition_N_1] 270.48 555.68
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_1-b_col_maj_False-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1] 48716.72 82563.34
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_2-b_col_maj_True-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1] 28658.18 24760.66
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_8-b_col_maj_True-c_col_maj_True-m_64-k_64-n_64-trace_size_0-partition_N_1] 7961.64 - -
test_gemm[M_384-K_1536-N_1792-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_32-k_48-n_64-trace_size_0-partition_N_1] 2246.30 5136.30
test_gemm[M_64-K_512-N_256-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_16-k_64-n_64-trace_size_0-partition_N_4] 2987.98 5873.74
test_gemm[M_896-K_1792-N_640-num_aie_columns_8-b_col_maj_False-c_col_maj_True-m_32-k_64-n_80-trace_size_0-partition_N_1] 1410.48 - -
iron/operators/gemv
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_gemv[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128] 0.23 0.08
test_gemv[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048] 11.38 3.72
test_gemv[M_2048-K_8192-num_aie_columns_2-tile_size_input_1-tile_size_output_1024] 23.90 5.81
test_gemv[M_2048-K_8192-num_aie_columns_4-tile_size_input_1-tile_size_output_512] 39.74 8.74
test_gemv[M_2048-K_8192-num_aie_columns_8-tile_size_input_1-tile_size_output_256] 43.95 - -
test_gemv[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024] 12.13 3.61
test_gemv[M_8192-K_2048-num_aie_columns_2-tile_size_input_4-tile_size_output_1024] 24.24 6.73
test_gemv[M_8192-K_2048-num_aie_columns_4-tile_size_input_4-tile_size_output_1024] 40.91 9.96
test_gemv[M_8192-K_2048-num_aie_columns_8-tile_size_input_4-tile_size_output_1024] 44.27 - -
test_gemv_gelu[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128] 0.20 -
test_gemv_gelu[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048] 12.84 -
test_gemv_gelu[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024] 12.65 -
iron/operators/layer_norm
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] 163.10 461.82
test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024] 182.10 319.52
test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] 163.12 419.28
test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512] 169.72 352.82
test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] 171.28 374.10
test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256] 174.66 768.88
test_layer_norm[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256] 189.58 - -
test_layer_norm[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128] 199.02 - -
iron/operators/leaky_relu
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.01] 162.30 399.78
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.1] 210.66 317.56
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.25] 167.22 393.48
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-alpha_0.01] 157.38 436.26
test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-alpha_0.01] 143.78 382.52
test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-alpha_0.01] 168.06 394.10
test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-alpha_0.01] 167.82 354.60
test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-alpha_0.01] 185.04 434.52
test_leaky_relu[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-alpha_0.01] 173.74 - -
test_leaky_relu[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128-alpha_0.01] 218.18 - -
iron/operators/mem_copy
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_mem_copy[input_length_2048-num_cores_1-num_channels_1-bypass_False-tile_size_2048] 133.56 320.42
test_mem_copy[input_length_2048-num_cores_16-num_channels_2-bypass_False-tile_size_128] 208.34 - -
test_mem_copy[input_length_2048-num_cores_2-num_channels_1-bypass_False-tile_size_1024] 160.76 435.24
test_mem_copy[input_length_2048-num_cores_2-num_channels_2-bypass_False-tile_size_1024] 160.48 404.36
test_mem_copy[input_length_2048-num_cores_4-num_channels_1-bypass_False-tile_size_512] 139.64 388.40
test_mem_copy[input_length_2048-num_cores_4-num_channels_2-bypass_False-tile_size_512] 170.94 423.30
test_mem_copy[input_length_2048-num_cores_8-num_channels_1-bypass_False-tile_size_256] 174.74 - -
test_mem_copy[input_length_2048-num_cores_8-num_channels_2-bypass_False-tile_size_256] 161.62 476.10
iron/operators/relu
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] 167.10 366.52
test_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024] 164.76 412.68
test_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] 178.84 336.84
test_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512] 188.46 393.38
test_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] 181.14 527.28
test_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256] 191.64 364.42
test_relu[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256] 203.30 - -
test_relu[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128] 235.30 - -
iron/operators/rms_norm
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_False] 163.52 477.68
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_True] 189.70 309.96
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_False] 170.38 348.64
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_True] 184.26 399.98
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_False] 171.52 433.54
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_True] 199.66 472.02
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_False] 175.20 395.62
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_True] 201.16 365.92
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_False] 172.56 513.12
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_True] 203.86 399.68
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_False] 192.48 600.06
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_True] 236.58 - -
test_rms_norm[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-weighted_False] 195.94 - -
test_rms_norm[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-weighted_True] 216.40 - -
test_rms_norm[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128-weighted_False] 210.04 - -
iron/operators/rope
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_1-method_type_0] 161.02 328.76
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_2-method_type_0] 160.88 568.56
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_4-method_type_0] 164.96 420.92
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_8-method_type_0] 189.06 - -
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_1-method_type_0] 169.90 410.40
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_2-method_type_0] 177.34 592.58
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_4-method_type_0] 181.70 442.78
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_8-method_type_0] 197.84 - -
iron/operators/sigmoid
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_sigmoid[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] 167.68 272.26
test_sigmoid[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024] 154.26 303.54
test_sigmoid[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] 164.74 292.54
test_sigmoid[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512] 141.58 286.18
test_sigmoid[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] 154.58 359.28
test_sigmoid[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256] 176.14 440.68
test_sigmoid[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256] 169.60 - -
test_sigmoid[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128] 217.20 - -
iron/operators/silu
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_silu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] 159.48 322.70
test_silu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] 174.32 324.44
test_silu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] 145.18 377.00
test_silu[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256] 163.98 - -
iron/operators/softmax
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_1024] 153.96 503.52
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_2048] 206.42 453.42
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_512] 206.48 493.96
iron/operators/swiglu_decode
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_swiglu_decode[embedding_dim_1024-hidden_dim_3584] 950.67 13848.60
test_swiglu_decode[embedding_dim_2048-hidden_dim_2048] 1003.76 15625.58
iron/operators/swiglu_prefill
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_swiglu_prefill[seq_len_256-embedding_dim_2048-hidden_dim_2048-prio_accuracy_False] 2241.40 20877.96
iron/operators/tanh
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_tanh[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] 153.38 338.82
test_tanh[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024] 182.04 324.20
test_tanh[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] 177.06 405.06
test_tanh[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512] 174.26 368.04
test_tanh[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] 184.22 433.34
test_tanh[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256] 189.20 452.56
test_tanh[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256] 206.56 - -
test_tanh[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128] 226.92 - -
iron/operators/transpose
Test Krackan Status Krackan Latency (mean) Phoenix Status Phoenix Latency (mean)
test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_1] 202.30 577.96
test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_2] 260.48 659.60
test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8-num_batches_1] 204.84 608.92
Krackan - Small

IRON

Tested on 2026_07_13_22_30_12 at commit fb80db1.

iron/operators/axpy
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_axpy[input_length_2048-num_aie_columns_1-tile_size_2048-scalar_factor_3.0]✅ 5/5139.100.09n/a
test_axpy[input_length_2048-num_aie_columns_2-tile_size_1024-scalar_factor_3.0]✅ 5/5188.500.07n/a
test_axpy[input_length_2048-num_aie_columns_4-tile_size_512-scalar_factor_3.0]✅ 5/5168.080.07n/a
test_axpy[input_length_2048-num_aie_columns_8-tile_size_256-scalar_factor_3.0]✅ 5/5183.620.07n/a
iron/operators/dequant
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_dequant[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-group_size_32]✅ 5/5178.700.03n/a
test_dequant[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-group_size_32]✅ 5/5154.000.03n/a
test_dequant[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-group_size_32]✅ 5/5177.760.03n/a
test_dequant[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-group_size_32]✅ 5/5173.760.03n/a
test_dequant[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-group_size_32]✅ 5/5179.140.03n/a
test_dequant[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-group_size_32]✅ 5/5229.880.02n/a
test_dequant[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-group_size_32]✅ 5/5250.580.02n/a
test_dequant[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128-group_size_32]✅ 5/5238.260.02n/a
iron/operators/elementwise_add
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_elementwise_add[input_length_2048-num_aie_columns_1-tile_size_2048]✅ 5/5175.360.07n/a
test_elementwise_add[input_length_2048-num_aie_columns_2-tile_size_1024]✅ 5/5182.200.07n/a
test_elementwise_add[input_length_2048-num_aie_columns_4-tile_size_512]✅ 5/5194.680.06n/a
test_elementwise_add[input_length_2048-num_aie_columns_8-tile_size_256]✅ 5/5192.960.06n/a
iron/operators/elementwise_mul
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_elementwise_mul[input_length_2048-num_aie_columns_1-tile_size_2048]✅ 5/5206.500.06n/a
test_elementwise_mul[input_length_2048-num_aie_columns_2-tile_size_1024]✅ 5/5157.220.08n/a
test_elementwise_mul[input_length_2048-num_aie_columns_4-tile_size_512]✅ 5/5181.880.07n/a
test_elementwise_mul[input_length_2048-num_aie_columns_8-tile_size_256]✅ 5/5207.820.06n/a
iron/operators/gelu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_gelu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5162.400.05n/a
test_gelu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5183.580.05n/a
test_gelu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5170.900.05n/a
test_gelu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5183.940.05n/a
test_gelu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5177.600.05n/a
test_gelu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5185.080.04n/a
test_gelu[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256]✅ 5/5200.140.04n/a
test_gelu[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128]✅ 5/5206.600.04n/a
iron/operators/gemm
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_gemm[M_1792-K_896-N_1152-num_aie_columns_8-b_col_maj_False-c_col_maj_True-m_64-k_32-n_48-trace_size_0-partition_N_1]✅ 5/52319.824.091608.11
test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_False-c_col_maj_False-m_48-k_96-n_16-trace_size_0-partition_N_1]✅ 5/5254.580.9440.04
test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_True-c_col_maj_True-m_48-k_96-n_16-trace_size_0-partition_N_1]✅ 5/5270.480.8937.90
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_1-b_col_maj_False-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]✅ 5/548716.720.52352.65
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_2-b_col_maj_True-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]✅ 5/528658.180.88599.49
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_8-b_col_maj_True-c_col_maj_True-m_64-k_64-n_64-trace_size_0-partition_N_1]✅ 5/57961.643.162158.83
test_gemm[M_384-K_1536-N_1792-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_32-k_48-n_64-trace_size_0-partition_N_1]✅ 5/52246.303.63950.96
test_gemm[M_64-K_512-N_256-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_16-k_64-n_64-trace_size_0-partition_N_4]✅ 5/52987.980.4322.96
test_gemm[M_896-K_1792-N_640-num_aie_columns_8-b_col_maj_False-c_col_maj_True-m_32-k_64-n_80-trace_size_0-partition_N_1]✅ 5/51410.485.011548.19
iron/operators/gemv
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_gemv[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]✅ 5/5n/a0.230.23
test_gemv[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]✅ 5/5n/a11.3811.37
test_gemv[M_2048-K_8192-num_aie_columns_2-tile_size_input_1-tile_size_output_1024]✅ 5/5n/a23.9023.88
test_gemv[M_2048-K_8192-num_aie_columns_4-tile_size_input_1-tile_size_output_512]✅ 5/5n/a39.7439.71
test_gemv[M_2048-K_8192-num_aie_columns_8-tile_size_input_1-tile_size_output_256]✅ 5/5n/a43.9543.93
test_gemv[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]✅ 5/5n/a12.1312.13
test_gemv[M_8192-K_2048-num_aie_columns_2-tile_size_input_4-tile_size_output_1024]✅ 5/5n/a24.2424.22
test_gemv[M_8192-K_2048-num_aie_columns_4-tile_size_input_4-tile_size_output_1024]✅ 5/5n/a40.9140.88
test_gemv[M_8192-K_2048-num_aie_columns_8-tile_size_input_4-tile_size_output_1024]✅ 5/5n/a44.2744.24
test_gemv_gelu[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]❌ 0/5n/a0.200.19
test_gemv_gelu[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]❌ 0/5n/a12.8412.83
test_gemv_gelu[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]❌ 0/5n/a12.6512.64
iron/operators/layer_norm
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5163.100.05n/a
test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5182.100.05n/a
test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5163.120.05n/a
test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5169.720.05n/a
test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5171.280.05n/a
test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5174.660.05n/a
test_layer_norm[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256]✅ 5/5189.580.04n/a
test_layer_norm[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128]✅ 5/5199.020.04n/a
iron/operators/leaky_relu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.01]✅ 5/5162.300.05n/a
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.1]✅ 5/5210.660.04n/a
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.25]✅ 5/5167.220.05n/a
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-alpha_0.01]✅ 5/5157.380.05n/a
test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-alpha_0.01]✅ 5/5143.780.06n/a
test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-alpha_0.01]✅ 5/5168.060.05n/a
test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-alpha_0.01]✅ 5/5167.820.05n/a
test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-alpha_0.01]✅ 5/5185.040.04n/a
test_leaky_relu[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-alpha_0.01]✅ 5/5173.740.05n/a
test_leaky_relu[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128-alpha_0.01]✅ 5/5218.180.04n/a
iron/operators/mem_copy
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_mem_copy[input_length_2048-num_cores_1-num_channels_1-bypass_False-tile_size_2048]✅ 5/5133.560.06n/a
test_mem_copy[input_length_2048-num_cores_16-num_channels_2-bypass_False-tile_size_128]✅ 5/5208.340.04n/a
test_mem_copy[input_length_2048-num_cores_2-num_channels_1-bypass_False-tile_size_1024]✅ 5/5160.760.05n/a
test_mem_copy[input_length_2048-num_cores_2-num_channels_2-bypass_False-tile_size_1024]✅ 5/5160.480.05n/a
test_mem_copy[input_length_2048-num_cores_4-num_channels_1-bypass_False-tile_size_512]✅ 5/5139.640.06n/a
test_mem_copy[input_length_2048-num_cores_4-num_channels_2-bypass_False-tile_size_512]✅ 5/5170.940.05n/a
test_mem_copy[input_length_2048-num_cores_8-num_channels_1-bypass_False-tile_size_256]✅ 5/5174.740.05n/a
test_mem_copy[input_length_2048-num_cores_8-num_channels_2-bypass_False-tile_size_256]✅ 5/5161.620.05n/a
iron/operators/relu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5167.100.05n/a
test_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5164.760.05n/a
test_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5178.840.05n/a
test_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5188.460.04n/a
test_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5181.140.05n/a
test_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5191.640.04n/a
test_relu[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256]✅ 5/5203.300.04n/a
test_relu[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128]✅ 5/5235.300.04n/a
iron/operators/rms_norm
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_False]✅ 5/5163.520.05n/a
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_True]✅ 5/5189.700.07n/a
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_False]✅ 5/5170.380.05n/a
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_True]✅ 5/5184.260.06n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_False]✅ 5/5171.520.05n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_True]✅ 5/5199.660.05n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_False]✅ 5/5175.200.05n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_True]✅ 5/5201.160.05n/a
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_False]✅ 5/5172.560.05n/a
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_True]✅ 5/5203.860.05n/a
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_False]✅ 5/5192.480.04n/a
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_True]✅ 5/5236.580.04n/a
test_rms_norm[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-weighted_False]✅ 5/5195.940.04n/a
test_rms_norm[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-weighted_True]✅ 5/5216.400.04n/a
test_rms_norm[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128-weighted_False]✅ 5/5210.040.04n/a
iron/operators/rope
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_1-method_type_0]✅ 5/5161.020.62n/a
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_2-method_type_0]✅ 5/5160.880.63n/a
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_4-method_type_0]✅ 5/5164.960.60n/a
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_8-method_type_0]✅ 5/5189.060.53n/a
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_1-method_type_0]✅ 5/5169.900.45n/a
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_2-method_type_0]✅ 5/5177.340.42n/a
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_4-method_type_0]✅ 5/5181.700.43n/a
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_8-method_type_0]✅ 5/5197.840.37n/a
iron/operators/sigmoid
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_sigmoid[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5167.680.05n/a
test_sigmoid[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5154.260.05n/a
test_sigmoid[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5164.740.05n/a
test_sigmoid[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5141.580.06n/a
test_sigmoid[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5154.580.05n/a
test_sigmoid[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5176.140.05n/a
test_sigmoid[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256]✅ 5/5169.600.05n/a
test_sigmoid[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128]✅ 5/5217.200.04n/a
iron/operators/silu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_silu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5159.480.05n/a
test_silu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5174.320.05n/a
test_silu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5145.180.06n/a
test_silu[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256]✅ 5/5163.980.05n/a
iron/operators/softmax
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_1024]✅ 5/5153.960.87n/a
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_2048]✅ 5/5206.420.66n/a
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5206.480.65n/a
iron/operators/swiglu_decode
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_swiglu_decode[embedding_dim_1024-hidden_dim_3584]✅ 5/5950.670.00n/a
test_swiglu_decode[embedding_dim_2048-hidden_dim_2048]✅ 5/51003.760.01n/a
iron/operators/swiglu_prefill
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_swiglu_prefill[seq_len_256-embedding_dim_2048-hidden_dim_2048-prio_accuracy_False]✅ 5/52241.400.94n/a
iron/operators/tanh
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_tanh[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5153.380.06n/a
test_tanh[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5182.040.05n/a
test_tanh[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5177.060.05n/a
test_tanh[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5174.260.05n/a
test_tanh[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5184.220.04n/a
test_tanh[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5189.200.04n/a
test_tanh[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256]✅ 5/5206.560.04n/a
test_tanh[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128]✅ 5/5226.920.04n/a
iron/operators/transpose
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_1]✅ 5/5202.302.71n/a
test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_2]✅ 5/5260.484.08n/a
test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8-num_batches_1]✅ 5/5204.842.66n/a

Trends:

IRON Trends

iron/operators/axpy

test_axpy[input_length_2048-num_aie_columns_1-tile_size_2048-scalar_factor_3.0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.11 (-15.07%)0.09 (+2.58%)0.08 (+5.33%)0.08 (+21.11%)0.01 (-43.13%)158.30 (-17.42%)139.10 (-5.69%)146.80 (-5.05%)113.60 (+17.72%)19.86 (-42.20%)
1f9e5a9 — 2026-07-13 18:35:330.13 (n/a)0.09 (n/a)0.08 (n/a)0.06 (n/a)0.02 (n/a)191.70 (n/a)147.50 (n/a)154.60 (n/a)96.50 (n/a)34.37 (n/a)

test_axpy[input_length_2048-num_aie_columns_2-tile_size_1024-scalar_factor_3.0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.09 (-18.79%)0.07 (-14.93%)0.07 (-18.19%)0.04 (-17.07%)0.02 (-21.75%)313.40 (+20.58%)188.50 (+17.07%)166.50 (+22.25%)140.20 (+23.20%)70.91 (+19.95%)
1f9e5a9 — 2026-07-13 18:35:330.11 (n/a)0.08 (n/a)0.09 (n/a)0.05 (n/a)0.02 (n/a)259.90 (n/a)161.02 (n/a)136.20 (n/a)113.80 (n/a)59.11 (n/a)

test_axpy[input_length_2048-num_aie_columns_4-tile_size_512-scalar_factor_3.0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.09 (-13.19%)0.07 (-3.51%)0.08 (-4.42%)0.06 (+67.70%)0.01 (-61.77%)189.20 (-40.37%)168.08 (-5.90%)161.00 (+4.61%)144.30 (+15.16%)19.72 (-74.90%)
1f9e5a9 — 2026-07-13 18:35:330.10 (n/a)0.08 (n/a)0.08 (n/a)0.04 (n/a)0.02 (n/a)317.30 (n/a)178.62 (n/a)153.90 (n/a)125.30 (n/a)78.60 (n/a)

test_axpy[input_length_2048-num_aie_columns_8-tile_size_256-scalar_factor_3.0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (-20.94%)0.07 (-14.28%)0.07 (-13.44%)0.06 (-9.10%)0.00 (-47.02%)208.60 (+10.02%)183.62 (+15.85%)180.70 (+15.54%)170.90 (+26.50%)14.62 (-26.11%)
1f9e5a9 — 2026-07-13 18:35:330.09 (n/a)0.08 (n/a)0.08 (n/a)0.06 (n/a)0.01 (n/a)189.60 (n/a)158.50 (n/a)156.40 (n/a)135.10 (n/a)19.79 (n/a)
iron/operators/dequant

test_dequant[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.04 (-12.12%)0.03 (-9.42%)0.03 (-12.39%)0.02 (+0.43%)0.01 (-16.41%)211.50 (-0.42%)178.70 (+9.55%)192.70 (+14.16%)120.60 (+13.77%)37.70 (-0.77%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)212.40 (n/a)163.12 (n/a)168.80 (n/a)106.00 (n/a)37.99 (n/a)

test_dequant[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.04 (-5.40%)0.03 (-2.45%)0.03 (-13.96%)0.03 (+8.08%)0.00 (-37.68%)178.10 (-7.48%)154.00 (+0.47%)158.50 (+16.20%)134.80 (+5.73%)18.86 (-40.32%)
1f9e5a9 — 2026-07-13 18:35:330.04 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)192.50 (n/a)153.28 (n/a)136.40 (n/a)127.50 (n/a)31.60 (n/a)

test_dequant[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.04 (+3.22%)0.03 (+27.27%)0.04 (+42.09%)0.02 (-12.31%)0.01 (+17.11%)332.20 (+14.04%)177.76 (-17.80%)144.90 (-29.63%)124.60 (-3.11%)87.56 (+31.92%)
1f9e5a9 — 2026-07-13 18:35:330.04 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)291.30 (n/a)216.26 (n/a)205.90 (n/a)128.60 (n/a)66.37 (n/a)

test_dequant[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.04 (+5.49%)0.03 (+3.92%)0.03 (+5.17%)0.02 (-4.96%)0.01 (+20.62%)218.60 (+5.20%)173.76 (-2.64%)185.60 (-4.92%)126.40 (-5.18%)37.13 (+18.45%)
1f9e5a9 — 2026-07-13 18:35:330.04 (n/a)0.03 (n/a)0.03 (n/a)0.03 (n/a)0.01 (n/a)207.80 (n/a)178.48 (n/a)195.20 (n/a)133.30 (n/a)31.34 (n/a)

test_dequant[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.03 (-4.39%)0.03 (+3.26%)0.03 (+3.50%)0.02 (+5.62%)0.00 (-23.98%)219.40 (-5.31%)179.14 (-3.92%)171.80 (-3.37%)165.40 (+4.62%)22.77 (-23.81%)
1f9e5a9 — 2026-07-13 18:35:330.03 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.00 (n/a)231.70 (n/a)186.44 (n/a)177.80 (n/a)158.10 (n/a)29.89 (n/a)

test_dequant[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.03 (-24.23%)0.02 (-18.86%)0.02 (-13.26%)0.02 (-25.04%)0.00 (-27.98%)280.30 (+33.41%)229.88 (+23.09%)220.40 (+15.27%)183.50 (+32.01%)36.59 (+30.84%)
1f9e5a9 — 2026-07-13 18:35:330.04 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)210.10 (n/a)186.76 (n/a)191.20 (n/a)139.00 (n/a)27.97 (n/a)

test_dequant[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.02 (-39.09%)0.02 (-28.93%)0.02 (-17.62%)0.01 (-39.28%)0.00 (-40.94%)353.80 (+64.63%)250.58 (+40.62%)226.30 (+21.41%)212.80 (+64.20%)58.25 (+65.42%)
1f9e5a9 — 2026-07-13 18:35:330.04 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)214.90 (n/a)178.20 (n/a)186.40 (n/a)129.60 (n/a)35.21 (n/a)

test_dequant[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.03 (-0.97%)0.02 (+9.43%)0.02 (+2.04%)0.02 (+14.29%)0.01 (-25.53%)344.00 (-12.49%)238.26 (-12.67%)219.30 (-1.97%)179.90 (+0.95%)62.47 (-33.64%)
1f9e5a9 — 2026-07-13 18:35:330.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)393.10 (n/a)272.84 (n/a)223.70 (n/a)178.20 (n/a)94.13 (n/a)
iron/operators/elementwise_add

test_elementwise_add[input_length_2048-num_aie_columns_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.09 (n/a)0.07 (n/a)0.07 (n/a)0.06 (n/a)0.01 (n/a)205.60 (n/a)175.36 (n/a)184.80 (n/a)140.10 (n/a)26.77 (n/a)

test_elementwise_add[input_length_2048-num_aie_columns_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.08 (n/a)0.07 (n/a)0.07 (n/a)0.05 (n/a)0.01 (n/a)227.80 (n/a)182.20 (n/a)184.40 (n/a)153.20 (n/a)29.37 (n/a)

test_elementwise_add[input_length_2048-num_aie_columns_4-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (n/a)0.06 (n/a)0.06 (n/a)0.06 (n/a)0.01 (n/a)214.80 (n/a)194.68 (n/a)194.40 (n/a)173.90 (n/a)17.75 (n/a)

test_elementwise_add[input_length_2048-num_aie_columns_8-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (n/a)0.06 (n/a)0.06 (n/a)0.06 (n/a)0.01 (n/a)211.90 (n/a)192.96 (n/a)201.90 (n/a)165.90 (n/a)19.28 (n/a)
iron/operators/elementwise_mul

test_elementwise_mul[input_length_2048-num_aie_columns_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (n/a)0.06 (n/a)0.06 (n/a)0.06 (n/a)0.01 (n/a)223.10 (n/a)206.50 (n/a)211.50 (n/a)174.30 (n/a)19.74 (n/a)

test_elementwise_mul[input_length_2048-num_aie_columns_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.11 (n/a)0.08 (n/a)0.07 (n/a)0.07 (n/a)0.02 (n/a)179.30 (n/a)157.22 (n/a)166.50 (n/a)114.20 (n/a)25.15 (n/a)

test_elementwise_mul[input_length_2048-num_aie_columns_4-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.09 (n/a)0.07 (n/a)0.06 (n/a)0.05 (n/a)0.02 (n/a)229.00 (n/a)181.88 (n/a)190.60 (n/a)132.60 (n/a)44.89 (n/a)

test_elementwise_mul[input_length_2048-num_aie_columns_8-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (n/a)0.06 (n/a)0.06 (n/a)0.05 (n/a)0.01 (n/a)248.10 (n/a)207.82 (n/a)207.00 (n/a)178.00 (n/a)30.78 (n/a)
iron/operators/gelu

test_gelu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)200.90 (n/a)162.40 (n/a)150.20 (n/a)126.10 (n/a)30.85 (n/a)

test_gelu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)211.40 (n/a)183.58 (n/a)167.90 (n/a)165.90 (n/a)23.15 (n/a)

test_gelu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)222.70 (n/a)170.90 (n/a)159.80 (n/a)128.60 (n/a)37.10 (n/a)

test_gelu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (n/a)0.05 (n/a)0.05 (n/a)0.03 (n/a)0.01 (n/a)282.60 (n/a)183.94 (n/a)162.60 (n/a)144.20 (n/a)57.67 (n/a)

test_gelu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.00 (n/a)204.10 (n/a)177.60 (n/a)170.60 (n/a)161.10 (n/a)18.84 (n/a)

test_gelu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.01 (n/a)203.50 (n/a)185.08 (n/a)194.00 (n/a)156.40 (n/a)20.37 (n/a)

test_gelu[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)253.90 (n/a)200.14 (n/a)188.30 (n/a)146.50 (n/a)42.50 (n/a)

test_gelu[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.04 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.00 (n/a)229.50 (n/a)206.60 (n/a)214.70 (n/a)184.90 (n/a)18.94 (n/a)
iron/operators/gemm

test_gemm[M_1792-K_896-N_1152-num_aie_columns_8-b_col_maj_False-c_col_maj_True-m_64-k_32-n_48-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:464.80 (+5.83%)4.09 (+0.23%)4.03 (-1.26%)3.71 (+7.72%)0.43 (+5.18%)2536.90 (-7.17%)2319.82 (-0.27%)2334.20 (+1.28%)1958.70 (-5.51%)228.45 (-9.34%)1888.65 (+5.83%)1608.11 (+0.23%)1584.84 (-1.26%)1458.22 (+7.72%)171.09 (+5.18%)
1f9e5a9 — 2026-07-13 18:35:334.54 (n/a)4.08 (n/a)4.08 (n/a)3.44 (n/a)0.41 (n/a)2732.80 (n/a)2326.16 (n/a)2304.70 (n/a)2072.90 (n/a)252.00 (n/a)1784.65 (n/a)1604.38 (n/a)1605.12 (n/a)1353.71 (n/a)162.66 (n/a)

test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_False-c_col_maj_False-m_48-k_96-n_16-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:461.24 (-3.45%)0.94 (-13.41%)1.01 (-7.97%)0.64 (-29.96%)0.28 (+65.62%)347.00 (+42.74%)254.58 (+22.35%)219.70 (+8.65%)178.80 (+3.53%)80.50 (+148.69%)52.77 (-3.45%)40.04 (-13.41%)42.96 (-7.97%)27.20 (-29.96%)11.83 (+65.62%)
1f9e5a9 — 2026-07-13 18:35:331.28 (n/a)1.08 (n/a)1.09 (n/a)0.91 (n/a)0.17 (n/a)243.10 (n/a)208.08 (n/a)202.20 (n/a)172.70 (n/a)32.37 (n/a)54.66 (n/a)46.24 (n/a)46.68 (n/a)38.83 (n/a)7.14 (n/a)

test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_True-c_col_maj_True-m_48-k_96-n_16-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:461.14 (-1.76%)0.89 (-1.04%)1.01 (+37.81%)0.59 (-18.52%)0.27 (+12.22%)378.10 (+22.72%)270.48 (+4.16%)218.30 (-27.45%)193.60 (+1.79%)89.91 (+43.46%)48.74 (-1.76%)37.90 (-1.04%)43.22 (+37.81%)24.96 (-18.52%)11.36 (+12.22%)
1f9e5a9 — 2026-07-13 18:35:331.16 (n/a)0.90 (n/a)0.74 (n/a)0.72 (n/a)0.24 (n/a)308.10 (n/a)259.68 (n/a)300.90 (n/a)190.20 (n/a)62.67 (n/a)49.62 (n/a)38.30 (n/a)31.36 (n/a)30.63 (n/a)10.13 (n/a)

test_gemm[M_2048-K_2048-N_2048-num_aie_columns_1-b_col_maj_False-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:460.52 (-0.14%)0.52 (-0.08%)0.52 (-0.12%)0.51 (-0.06%)0.00 (-15.86%)48884.80 (+0.06%)48716.72 (+0.08%)48712.60 (+0.12%)48613.30 (+0.14%)103.49 (-15.67%)353.40 (-0.14%)352.65 (-0.08%)352.68 (-0.12%)351.44 (-0.06%)0.75 (-15.86%)
1f9e5a9 — 2026-07-13 18:35:330.52 (n/a)0.52 (n/a)0.52 (n/a)0.52 (n/a)0.00 (n/a)48855.70 (n/a)48678.68 (n/a)48655.50 (n/a)48546.90 (n/a)122.72 (n/a)353.88 (n/a)352.93 (n/a)353.09 (n/a)351.64 (n/a)0.89 (n/a)

test_gemm[M_2048-K_2048-N_2048-num_aie_columns_2-b_col_maj_True-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:460.89 (-0.84%)0.88 (-0.64%)0.88 (-0.77%)0.87 (-0.44%)0.01 (-22.96%)28815.50 (+0.44%)28658.18 (+0.64%)28738.60 (+0.78%)28427.30 (+0.85%)163.26 (-21.99%)604.34 (-0.84%)599.49 (-0.64%)597.80 (-0.77%)596.20 (-0.44%)3.42 (-22.96%)
1f9e5a9 — 2026-07-13 18:35:330.89 (n/a)0.88 (n/a)0.88 (n/a)0.88 (n/a)0.01 (n/a)28688.90 (n/a)28476.06 (n/a)28516.40 (n/a)28188.10 (n/a)209.28 (n/a)609.47 (n/a)603.34 (n/a)602.46 (n/a)598.83 (n/a)4.44 (n/a)

test_gemm[M_2048-K_2048-N_2048-num_aie_columns_8-b_col_maj_True-c_col_maj_True-m_64-k_64-n_64-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:463.29 (+1.21%)3.16 (+0.03%)3.14 (-0.17%)3.09 (-0.30%)0.08 (+36.58%)8150.40 (+0.31%)7961.64 (-0.01%)8014.50 (+0.17%)7650.70 (-1.20%)188.67 (+35.14%)2245.52 (+1.21%)2158.83 (+0.03%)2143.61 (-0.17%)2107.87 (-0.30%)52.22 (+36.58%)
1f9e5a9 — 2026-07-13 18:35:333.25 (n/a)3.16 (n/a)3.15 (n/a)3.10 (n/a)0.06 (n/a)8125.50 (n/a)7962.46 (n/a)8000.50 (n/a)7743.40 (n/a)139.62 (n/a)2218.65 (n/a)2158.14 (n/a)2147.35 (n/a)2114.31 (n/a)38.23 (n/a)

test_gemm[M_384-K_1536-N_1792-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_32-k_48-n_64-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:464.06 (+3.19%)3.63 (+2.74%)3.70 (+3.70%)2.99 (+3.56%)0.40 (-1.11%)2694.00 (-3.44%)2246.30 (-2.75%)2176.30 (-3.56%)1983.50 (-3.09%)268.51 (-6.94%)1065.77 (+3.19%)950.96 (+2.74%)971.34 (+3.70%)784.68 (+3.56%)103.88 (-1.11%)
1f9e5a9 — 2026-07-13 18:35:333.94 (n/a)3.53 (n/a)3.57 (n/a)2.89 (n/a)0.40 (n/a)2790.00 (n/a)2309.92 (n/a)2256.70 (n/a)2046.70 (n/a)288.52 (n/a)1032.84 (n/a)925.60 (n/a)936.73 (n/a)757.68 (n/a)105.05 (n/a)

test_gemm[M_64-K_512-N_256-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_16-k_64-n_64-trace_size_0-partition_N_4]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:460.50 (+0.56%)0.43 (+12.39%)0.42 (+13.03%)0.35 (+13.26%)0.07 (-2.27%)3558.20 (-11.71%)2987.98 (-11.33%)2979.60 (-11.53%)2480.10 (-0.55%)491.66 (-12.38%)27.06 (+0.56%)22.96 (+12.39%)22.52 (+13.03%)18.86 (+13.26%)3.78 (-2.27%)
1f9e5a9 — 2026-07-13 18:35:330.50 (n/a)0.38 (n/a)0.37 (n/a)0.31 (n/a)0.07 (n/a)4030.20 (n/a)3369.80 (n/a)3368.00 (n/a)2493.90 (n/a)561.13 (n/a)26.91 (n/a)20.42 (n/a)19.93 (n/a)16.65 (n/a)3.87 (n/a)

test_gemm[M_896-K_1792-N_640-num_aie_columns_8-b_col_maj_False-c_col_maj_True-m_32-k_64-n_80-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:466.48 (+21.21%)5.01 (+0.18%)4.67 (-6.75%)3.68 (-21.02%)1.38 (+403.05%)1806.70 (+26.62%)1410.48 (+5.80%)1425.90 (+7.23%)1026.60 (-17.50%)378.40 (+415.32%)2001.95 (+21.21%)1548.19 (+0.18%)1441.34 (-6.75%)1137.57 (-21.02%)427.41 (+403.05%)
1f9e5a9 — 2026-07-13 18:35:335.35 (n/a)5.00 (n/a)5.00 (n/a)4.66 (n/a)0.27 (n/a)1426.90 (n/a)1333.12 (n/a)1329.70 (n/a)1244.30 (n/a)73.43 (n/a)1651.64 (n/a)1545.37 (n/a)1545.60 (n/a)1440.31 (n/a)84.96 (n/a)
iron/operators/gemv

test_gemv[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:460.27 (n/a)0.23 (n/a)0.24 (n/a)0.19 (n/a)0.04 (n/a)0.27 (n/a)0.23 (n/a)0.24 (n/a)0.19 (n/a)0.03 (n/a)

test_gemv[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4612.43 (n/a)11.38 (n/a)11.16 (n/a)10.67 (n/a)0.79 (n/a)12.43 (n/a)11.37 (n/a)11.15 (n/a)10.66 (n/a)0.79 (n/a)

test_gemv[M_2048-K_8192-num_aie_columns_2-tile_size_input_1-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4624.95 (-0.40%)23.90 (+5.78%)24.05 (+3.77%)22.38 (+26.56%)0.93 (-67.21%)24.94 (-0.40%)23.88 (+5.78%)24.04 (+3.77%)22.37 (+26.56%)0.93 (-67.21%)
1f9e5a9 — 2026-07-13 18:35:3325.05 (n/a)22.59 (n/a)23.18 (n/a)17.69 (n/a)2.85 (n/a)25.04 (n/a)22.58 (n/a)23.16 (n/a)17.68 (n/a)2.85 (n/a)

test_gemv[M_2048-K_8192-num_aie_columns_4-tile_size_input_1-tile_size_output_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4640.61 (-5.83%)39.74 (+0.03%)39.82 (+0.95%)38.36 (+1.58%)0.84 (-60.57%)40.58 (-5.83%)39.71 (+0.03%)39.80 (+0.95%)38.34 (+1.58%)0.83 (-60.57%)
1f9e5a9 — 2026-07-13 18:35:3343.12 (n/a)39.72 (n/a)39.45 (n/a)37.76 (n/a)2.12 (n/a)43.09 (n/a)39.70 (n/a)39.42 (n/a)37.74 (n/a)2.12 (n/a)

test_gemv[M_2048-K_8192-num_aie_columns_8-tile_size_input_1-tile_size_output_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4645.64 (+1.56%)43.95 (+2.66%)44.24 (+3.17%)42.48 (+3.40%)1.34 (-16.85%)45.61 (+1.56%)43.93 (+2.66%)44.21 (+3.17%)42.46 (+3.40%)1.34 (-16.85%)
1f9e5a9 — 2026-07-13 18:35:3344.94 (n/a)42.82 (n/a)42.88 (n/a)41.09 (n/a)1.61 (n/a)44.91 (n/a)42.79 (n/a)42.85 (n/a)41.06 (n/a)1.61 (n/a)

test_gemv[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4613.49 (n/a)12.13 (n/a)12.54 (n/a)10.54 (n/a)1.37 (n/a)13.48 (n/a)12.13 (n/a)12.53 (n/a)10.54 (n/a)1.37 (n/a)

test_gemv[M_8192-K_2048-num_aie_columns_2-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4625.06 (+0.31%)24.24 (+1.57%)24.24 (-0.54%)23.50 (+11.26%)0.57 (-63.69%)25.04 (+0.31%)24.22 (+1.57%)24.22 (-0.54%)23.48 (+11.26%)0.57 (-63.69%)
1f9e5a9 — 2026-07-13 18:35:3324.98 (n/a)23.86 (n/a)24.37 (n/a)21.12 (n/a)1.58 (n/a)24.96 (n/a)23.85 (n/a)24.35 (n/a)21.10 (n/a)1.58 (n/a)

test_gemv[M_8192-K_2048-num_aie_columns_4-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4642.37 (+4.81%)40.91 (+4.70%)40.33 (+3.37%)39.65 (+7.73%)1.25 (-15.61%)42.34 (+4.81%)40.88 (+4.70%)40.31 (+3.37%)39.62 (+7.73%)1.25 (-15.61%)
1f9e5a9 — 2026-07-13 18:35:3340.42 (n/a)39.07 (n/a)39.01 (n/a)36.80 (n/a)1.48 (n/a)40.40 (n/a)39.05 (n/a)38.99 (n/a)36.78 (n/a)1.48 (n/a)

test_gemv[M_8192-K_2048-num_aie_columns_8-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4646.67 (+3.68%)44.27 (+2.27%)43.72 (-0.24%)42.28 (+2.09%)1.97 (+36.68%)46.64 (+3.68%)44.24 (+2.27%)43.69 (-0.24%)42.25 (+2.09%)1.97 (+36.68%)
1f9e5a9 — 2026-07-13 18:35:3345.01 (n/a)43.29 (n/a)43.82 (n/a)41.41 (n/a)1.44 (n/a)44.98 (n/a)43.26 (n/a)43.80 (n/a)41.38 (n/a)1.44 (n/a)

test_gemv_gelu[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:460.25 (n/a)0.20 (n/a)0.18 (n/a)0.17 (n/a)0.03 (n/a)0.25 (n/a)0.19 (n/a)0.18 (n/a)0.17 (n/a)0.03 (n/a)

test_gemv_gelu[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4613.31 (n/a)12.84 (n/a)13.04 (n/a)12.20 (n/a)0.49 (n/a)13.30 (n/a)12.83 (n/a)13.03 (n/a)12.19 (n/a)0.49 (n/a)

test_gemv_gelu[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4613.27 (n/a)12.65 (n/a)12.91 (n/a)11.66 (n/a)0.72 (n/a)13.26 (n/a)12.64 (n/a)12.90 (n/a)11.66 (n/a)0.72 (n/a)
iron/operators/layer_norm

test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)205.00 (n/a)163.10 (n/a)170.00 (n/a)120.60 (n/a)34.38 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (n/a)0.05 (n/a)0.04 (n/a)0.04 (n/a)0.01 (n/a)203.70 (n/a)182.10 (n/a)183.70 (n/a)149.10 (n/a)20.51 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (n/a)0.05 (n/a)0.05 (n/a)0.05 (n/a)0.01 (n/a)180.80 (n/a)163.12 (n/a)166.20 (n/a)127.60 (n/a)21.17 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.00 (n/a)192.40 (n/a)169.72 (n/a)170.80 (n/a)148.90 (n/a)16.50 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.00 (n/a)189.80 (n/a)171.28 (n/a)171.90 (n/a)157.60 (n/a)12.04 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.00 (n/a)198.20 (n/a)174.66 (n/a)173.10 (n/a)153.40 (n/a)16.04 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.01 (n/a)230.70 (n/a)189.58 (n/a)185.00 (n/a)154.20 (n/a)28.20 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.00 (n/a)224.10 (n/a)199.02 (n/a)204.00 (n/a)178.00 (n/a)19.99 (n/a)
iron/operators/leaky_relu

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (+6.89%)0.05 (+16.02%)0.05 (+20.76%)0.04 (+21.48%)0.01 (-29.34%)189.90 (-17.69%)162.30 (-15.36%)161.70 (-17.20%)136.60 (-6.44%)19.27 (-46.08%)
1f9e5a9 — 2026-07-13 18:35:330.06 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.01 (n/a)230.70 (n/a)191.76 (n/a)195.30 (n/a)146.00 (n/a)35.73 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (-14.11%)0.04 (-16.86%)0.04 (-12.28%)0.02 (-43.44%)0.01 (+23.26%)345.80 (+76.79%)210.66 (+27.73%)192.00 (+14.01%)145.80 (+16.45%)79.60 (+157.36%)
1f9e5a9 — 2026-07-13 18:35:330.07 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)195.60 (n/a)164.92 (n/a)168.40 (n/a)125.20 (n/a)30.93 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.25]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (+6.96%)0.05 (+2.69%)0.05 (+1.12%)0.04 (-2.62%)0.00 (+69.62%)192.20 (+2.67%)167.22 (-2.21%)165.00 (-1.08%)153.20 (-6.47%)15.32 (+62.45%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.00 (n/a)187.20 (n/a)171.00 (n/a)166.80 (n/a)163.80 (n/a)9.43 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (+21.96%)0.05 (+16.90%)0.05 (+4.02%)0.05 (+7.04%)0.01 (+106.83%)180.80 (-6.56%)157.38 (-12.40%)179.10 (-3.86%)120.50 (-17.97%)30.79 (+64.00%)
1f9e5a9 — 2026-07-13 18:35:330.06 (n/a)0.05 (n/a)0.04 (n/a)0.04 (n/a)0.01 (n/a)193.50 (n/a)179.66 (n/a)186.30 (n/a)146.90 (n/a)18.78 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (+31.20%)0.06 (+25.12%)0.06 (+27.33%)0.04 (+5.28%)0.01 (+96.47%)186.00 (-5.01%)143.78 (-18.94%)139.20 (-21.44%)120.00 (-23.81%)25.32 (+45.20%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.00 (n/a)195.80 (n/a)177.38 (n/a)177.20 (n/a)157.50 (n/a)17.44 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (+15.29%)0.05 (+11.71%)0.05 (+18.33%)0.04 (+9.93%)0.01 (+35.05%)199.80 (-9.02%)168.06 (-10.04%)158.60 (-15.46%)142.50 (-13.22%)23.32 (+7.65%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.00 (n/a)219.60 (n/a)186.82 (n/a)187.60 (n/a)164.20 (n/a)21.67 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (+6.92%)0.05 (-4.91%)0.05 (-9.20%)0.04 (-4.77%)0.01 (+25.64%)201.80 (+4.99%)167.82 (+6.13%)177.50 (+10.18%)126.20 (-6.45%)28.86 (+23.44%)
1f9e5a9 — 2026-07-13 18:35:330.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)192.20 (n/a)158.12 (n/a)161.10 (n/a)134.90 (n/a)23.38 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (+8.79%)0.04 (+9.69%)0.05 (+13.95%)0.04 (+7.44%)0.01 (+14.59%)212.70 (-6.91%)185.04 (-8.69%)176.60 (-12.23%)159.20 (-8.03%)22.84 (-1.43%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.00 (n/a)228.50 (n/a)202.64 (n/a)201.20 (n/a)173.10 (n/a)23.17 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (-4.36%)0.05 (+0.23%)0.05 (+0.27%)0.04 (+9.93%)0.01 (-20.28%)202.60 (-9.03%)173.74 (-1.45%)176.70 (-0.23%)133.90 (+4.53%)25.94 (-24.04%)
1f9e5a9 — 2026-07-13 18:35:330.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)222.70 (n/a)176.30 (n/a)177.10 (n/a)128.10 (n/a)34.15 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (+1.59%)0.04 (-5.02%)0.04 (-4.07%)0.03 (-7.32%)0.01 (+7.93%)249.50 (+7.92%)218.18 (+5.53%)223.30 (+4.25%)171.20 (-1.55%)28.60 (+9.90%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.01 (n/a)231.20 (n/a)206.74 (n/a)214.20 (n/a)173.90 (n/a)26.03 (n/a)
iron/operators/mem_copy

test_mem_copy[input_length_2048-num_cores_1-num_channels_1-bypass_False-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (+8.67%)0.06 (+22.70%)0.06 (+32.18%)0.05 (+29.68%)0.01 (-19.64%)162.50 (-22.88%)133.56 (-19.53%)126.60 (-24.37%)119.70 (-7.92%)16.99 (-42.25%)
1f9e5a9 — 2026-07-13 18:35:330.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)210.70 (n/a)165.98 (n/a)167.40 (n/a)130.00 (n/a)29.43 (n/a)

test_mem_copy[input_length_2048-num_cores_16-num_channels_2-bypass_False-tile_size_128]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (-19.56%)0.04 (-2.45%)0.04 (+0.43%)0.03 (+0.50%)0.01 (-36.90%)312.30 (-0.51%)208.34 (-1.74%)185.60 (-0.43%)175.50 (+24.29%)58.27 (-19.92%)
1f9e5a9 — 2026-07-13 18:35:330.06 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)313.90 (n/a)212.04 (n/a)186.40 (n/a)141.20 (n/a)72.76 (n/a)

test_mem_copy[input_length_2048-num_cores_2-num_channels_1-bypass_False-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (+2.48%)0.05 (+14.06%)0.06 (+26.90%)0.04 (+0.03%)0.01 (+21.06%)230.00 (+0.00%)160.76 (-10.82%)134.40 (-21.17%)124.00 (-2.44%)46.37 (+15.82%)
1f9e5a9 — 2026-07-13 18:35:330.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)230.00 (n/a)180.26 (n/a)170.50 (n/a)127.10 (n/a)40.03 (n/a)

test_mem_copy[input_length_2048-num_cores_2-num_channels_2-bypass_False-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (-0.55%)0.05 (+12.32%)0.06 (+25.35%)0.04 (+18.13%)0.01 (-11.46%)227.50 (-15.33%)160.48 (-12.59%)142.20 (-20.20%)124.20 (+0.57%)40.36 (-23.70%)
1f9e5a9 — 2026-07-13 18:35:330.07 (n/a)0.05 (n/a)0.05 (n/a)0.03 (n/a)0.01 (n/a)268.70 (n/a)183.60 (n/a)178.20 (n/a)123.50 (n/a)52.89 (n/a)

test_mem_copy[input_length_2048-num_cores_4-num_channels_1-bypass_False-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (+10.52%)0.06 (+13.25%)0.06 (+13.06%)0.05 (+9.92%)0.01 (+15.30%)165.50 (-9.02%)139.64 (-11.62%)134.80 (-11.55%)122.80 (-9.51%)17.05 (-5.41%)
1f9e5a9 — 2026-07-13 18:35:330.06 (n/a)0.05 (n/a)0.05 (n/a)0.05 (n/a)0.01 (n/a)181.90 (n/a)158.00 (n/a)152.40 (n/a)135.70 (n/a)18.03 (n/a)

test_mem_copy[input_length_2048-num_cores_4-num_channels_2-bypass_False-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (-2.93%)0.05 (+1.60%)0.05 (-2.36%)0.03 (-10.07%)0.01 (+12.37%)261.70 (+11.22%)170.94 (+0.36%)165.40 (+2.41%)127.80 (+2.98%)53.98 (+27.98%)
1f9e5a9 — 2026-07-13 18:35:330.07 (n/a)0.05 (n/a)0.05 (n/a)0.03 (n/a)0.01 (n/a)235.30 (n/a)170.32 (n/a)161.50 (n/a)124.10 (n/a)42.18 (n/a)

test_mem_copy[input_length_2048-num_cores_8-num_channels_1-bypass_False-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (+2.64%)0.05 (+1.79%)0.05 (+7.51%)0.04 (-15.94%)0.01 (+78.94%)228.70 (+18.93%)174.74 (-0.17%)158.20 (-6.94%)152.30 (-2.62%)31.53 (+107.60%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.00 (n/a)192.30 (n/a)175.04 (n/a)170.00 (n/a)156.40 (n/a)15.19 (n/a)

test_mem_copy[input_length_2048-num_cores_8-num_channels_2-bypass_False-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (+17.79%)0.05 (+18.37%)0.05 (+7.00%)0.04 (+54.37%)0.01 (-6.25%)211.30 (-35.20%)161.62 (-18.90%)159.10 (-6.58%)128.60 (-15.12%)34.15 (-52.41%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.04 (n/a)0.05 (n/a)0.03 (n/a)0.01 (n/a)326.10 (n/a)199.28 (n/a)170.30 (n/a)151.50 (n/a)71.76 (n/a)
iron/operators/mha

test_mha[seq_len_16384-dim_64-num_heads_1-num_pipelines_8-num_kv_heads_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
1f9e5a9 — 2026-07-13 18:35:330.21 (+0.21%)0.21 (+0.15%)0.21 (+0.30%)0.20 (-0.21%)0.00 (+162.02%)41048.30 (+0.21%)40864.76 (-0.15%)40818.90 (-0.30%)40772.70 (-0.21%)109.15 (+161.97%)
573678d — 2026-06-29 18:46:460.21 (n/a)0.20 (n/a)0.20 (n/a)0.20 (n/a)0.00 (n/a)40961.20 (n/a)40924.92 (n/a)40941.50 (n/a)40859.60 (n/a)41.66 (n/a)
iron/operators/rms_norm

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (+12.05%)0.05 (+12.90%)0.05 (+15.52%)0.05 (+20.42%)0.01 (-15.08%)178.70 (-16.96%)163.52 (-12.03%)167.70 (-13.47%)140.10 (-10.71%)16.13 (-35.88%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.01 (n/a)215.20 (n/a)185.88 (n/a)193.80 (n/a)156.90 (n/a)25.15 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.09 (-11.59%)0.07 (-13.04%)0.07 (-12.89%)0.05 (-25.20%)0.02 (+27.86%)243.90 (+33.64%)189.70 (+17.94%)184.70 (+14.79%)141.70 (+13.18%)43.45 (+97.60%)
1f9e5a9 — 2026-07-13 18:35:330.10 (n/a)0.08 (n/a)0.08 (n/a)0.07 (n/a)0.01 (n/a)182.50 (n/a)160.84 (n/a)160.90 (n/a)125.20 (n/a)21.99 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (+5.91%)0.05 (+10.92%)0.06 (+36.65%)0.03 (-39.13%)0.02 (+97.08%)317.90 (+64.29%)170.38 (-0.25%)132.70 (-26.85%)127.20 (-5.57%)82.67 (+211.53%)
1f9e5a9 — 2026-07-13 18:35:330.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)193.50 (n/a)170.80 (n/a)181.40 (n/a)134.70 (n/a)26.54 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (-30.82%)0.06 (-14.25%)0.06 (-0.23%)0.05 (-19.13%)0.01 (-50.10%)220.80 (+23.63%)184.26 (+15.00%)170.70 (+0.23%)168.00 (+44.58%)23.01 (-10.19%)
1f9e5a9 — 2026-07-13 18:35:330.09 (n/a)0.07 (n/a)0.06 (n/a)0.06 (n/a)0.01 (n/a)178.60 (n/a)160.22 (n/a)170.30 (n/a)116.20 (n/a)25.62 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (-16.92%)0.05 (-0.19%)0.05 (+2.98%)0.04 (+13.77%)0.01 (-48.58%)189.30 (-12.12%)171.52 (-3.46%)179.70 (-2.92%)134.00 (+20.29%)22.32 (-43.23%)
1f9e5a9 — 2026-07-13 18:35:330.07 (n/a)0.05 (n/a)0.04 (n/a)0.04 (n/a)0.01 (n/a)215.40 (n/a)177.66 (n/a)185.10 (n/a)111.40 (n/a)39.31 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (-38.41%)0.05 (-23.78%)0.06 (-7.07%)0.04 (-29.57%)0.01 (-43.37%)254.00 (+41.98%)199.66 (+29.95%)178.40 (+7.60%)166.00 (+62.27%)40.40 (+34.41%)
1f9e5a9 — 2026-07-13 18:35:330.10 (n/a)0.07 (n/a)0.06 (n/a)0.06 (n/a)0.02 (n/a)178.90 (n/a)153.64 (n/a)165.80 (n/a)102.30 (n/a)30.06 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (-9.42%)0.05 (-4.04%)0.05 (+10.54%)0.04 (+10.57%)0.01 (-34.51%)213.00 (-9.55%)175.20 (-0.44%)167.40 (-9.56%)125.50 (+10.38%)36.47 (-32.41%)
1f9e5a9 — 2026-07-13 18:35:330.07 (n/a)0.05 (n/a)0.04 (n/a)0.03 (n/a)0.02 (n/a)235.50 (n/a)175.98 (n/a)185.10 (n/a)113.70 (n/a)53.96 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (-12.86%)0.05 (-3.60%)0.05 (-3.42%)0.04 (+5.97%)0.01 (-41.05%)223.20 (-5.62%)201.16 (+1.90%)199.00 (+3.54%)165.70 (+14.75%)22.89 (-35.53%)
1f9e5a9 — 2026-07-13 18:35:330.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)236.50 (n/a)197.40 (n/a)192.20 (n/a)144.40 (n/a)35.51 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.06 (+16.01%)0.05 (+19.33%)0.05 (+23.43%)0.04 (+10.33%)0.01 (+29.75%)215.30 (-9.35%)172.56 (-15.43%)165.70 (-18.97%)129.40 (-13.85%)36.28 (+4.03%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)237.50 (n/a)204.04 (n/a)204.50 (n/a)150.20 (n/a)34.87 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (+3.19%)0.05 (+0.57%)0.04 (-0.69%)0.04 (-0.48%)0.00 (+22.61%)220.20 (+0.46%)203.86 (-0.31%)210.10 (+0.67%)171.90 (-3.10%)19.94 (+20.23%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.05 (n/a)0.04 (n/a)0.04 (n/a)0.00 (n/a)219.20 (n/a)204.50 (n/a)208.70 (n/a)177.40 (n/a)16.58 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.07 (+61.39%)0.04 (+19.84%)0.04 (+5.95%)0.03 (+1.66%)0.01 (+306.74%)235.50 (-1.67%)192.48 (-12.72%)206.70 (-5.62%)124.20 (-38.05%)43.04 (+137.40%)
1f9e5a9 — 2026-07-13 18:35:330.04 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.00 (n/a)239.50 (n/a)220.54 (n/a)219.00 (n/a)200.50 (n/a)18.13 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.04 (-17.09%)0.04 (-10.13%)0.04 (-6.10%)0.03 (-4.57%)0.01 (-31.67%)319.40 (+4.79%)236.58 (+9.42%)223.60 (+6.48%)205.70 (+20.57%)47.31 (-12.17%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)304.80 (n/a)216.22 (n/a)210.00 (n/a)170.60 (n/a)53.86 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (+7.09%)0.04 (+8.14%)0.04 (+7.65%)0.03 (+1.19%)0.00 (+4.00%)234.50 (-1.14%)195.94 (-7.53%)196.60 (-7.13%)172.50 (-6.61%)24.11 (-4.12%)
1f9e5a9 — 2026-07-13 18:35:330.04 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.00 (n/a)237.20 (n/a)211.90 (n/a)211.70 (n/a)184.70 (n/a)25.15 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_8-num_channels_1-tile_size_256-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (+8.80%)0.04 (+4.10%)0.04 (+6.12%)0.03 (+0.99%)0.00 (+58.57%)250.40 (-0.95%)216.40 (-3.39%)207.10 (-5.78%)192.20 (-8.08%)24.68 (+43.30%)
1f9e5a9 — 2026-07-13 18:35:330.04 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.00 (n/a)252.80 (n/a)224.00 (n/a)219.80 (n/a)209.10 (n/a)17.23 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_8-num_channels_2-tile_size_128-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.05 (-0.38%)0.04 (-3.34%)0.04 (-0.43%)0.03 (-4.08%)0.00 (+2.39%)238.60 (+4.24%)210.04 (+3.52%)207.30 (+0.39%)180.10 (+0.39%)21.76 (+7.70%)
1f9e5a9 — 2026-07-13 18:35:330.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.00 (n/a)228.90 (n/a)202.90 (n/a)206.50 (n/a)179.40 (n/a)20.20 (n/a)
iron/operators/rope

test_rope[rows_32-cols_512-angle_rows_32-aie_columns_1-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.71 (+23.78%)0.62 (+29.66%)0.60 (+26.81%)0.53 (+37.53%)0.07 (-14.81%)184.00 (-27.30%)161.02 (-23.81%)164.90 (-21.10%)137.90 (-19.17%)17.05 (-50.53%)
1f9e5a9 — 2026-07-13 18:35:330.58 (n/a)0.48 (n/a)0.47 (n/a)0.39 (n/a)0.08 (n/a)253.10 (n/a)211.34 (n/a)209.00 (n/a)170.60 (n/a)34.47 (n/a)

test_rope[rows_32-cols_512-angle_rows_32-aie_columns_2-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.77 (+2.42%)0.63 (+4.47%)0.58 (+6.65%)0.51 (-4.95%)0.11 (+23.21%)194.20 (+5.20%)160.88 (-3.44%)168.40 (-6.24%)128.20 (-2.36%)28.10 (+23.79%)
1f9e5a9 — 2026-07-13 18:35:330.75 (n/a)0.60 (n/a)0.55 (n/a)0.53 (n/a)0.09 (n/a)184.60 (n/a)166.62 (n/a)179.60 (n/a)131.30 (n/a)22.70 (n/a)

test_rope[rows_32-cols_512-angle_rows_32-aie_columns_4-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.69 (+2.60%)0.60 (-0.87%)0.56 (-11.20%)0.52 (-4.15%)0.08 (+44.57%)188.50 (+4.37%)164.96 (+1.61%)175.60 (+12.56%)141.80 (-2.48%)21.44 (+41.24%)
1f9e5a9 — 2026-07-13 18:35:330.68 (n/a)0.61 (n/a)0.63 (n/a)0.54 (n/a)0.06 (n/a)180.60 (n/a)162.34 (n/a)156.00 (n/a)145.40 (n/a)15.18 (n/a)

test_rope[rows_32-cols_512-angle_rows_32-aie_columns_8-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.65 (+0.12%)0.53 (+4.78%)0.50 (+5.86%)0.42 (-8.03%)0.09 (+13.17%)232.20 (+8.71%)189.06 (-3.96%)194.80 (-5.53%)151.60 (-0.13%)31.57 (+24.17%)
1f9e5a9 — 2026-07-13 18:35:330.65 (n/a)0.51 (n/a)0.48 (n/a)0.46 (n/a)0.08 (n/a)213.60 (n/a)196.86 (n/a)206.20 (n/a)151.80 (n/a)25.42 (n/a)

test_rope[rows_32-cols_512-angle_rows_8-aie_columns_1-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.57 (+5.21%)0.45 (-7.24%)0.44 (-9.70%)0.35 (-17.68%)0.08 (+73.23%)208.80 (+21.47%)169.90 (+9.91%)169.40 (+10.79%)130.00 (-4.97%)30.86 (+97.94%)
1f9e5a9 — 2026-07-13 18:35:330.54 (n/a)0.48 (n/a)0.48 (n/a)0.43 (n/a)0.05 (n/a)171.90 (n/a)154.58 (n/a)152.90 (n/a)136.80 (n/a)15.59 (n/a)

test_rope[rows_32-cols_512-angle_rows_8-aie_columns_2-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.44 (-8.86%)0.42 (+5.25%)0.40 (+3.97%)0.40 (+15.85%)0.02 (-65.28%)184.50 (-13.66%)177.34 (-6.14%)182.10 (-3.85%)166.50 (+9.76%)7.91 (-66.60%)
1f9e5a9 — 2026-07-13 18:35:330.49 (n/a)0.40 (n/a)0.39 (n/a)0.34 (n/a)0.05 (n/a)213.70 (n/a)188.94 (n/a)189.40 (n/a)151.70 (n/a)23.68 (n/a)

test_rope[rows_32-cols_512-angle_rows_8-aie_columns_4-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.56 (-1.35%)0.43 (-4.71%)0.44 (-9.05%)0.31 (+9.85%)0.11 (-5.20%)235.20 (-8.98%)181.70 (+4.03%)167.50 (+9.91%)132.70 (+1.38%)46.73 (-10.39%)
1f9e5a9 — 2026-07-13 18:35:330.56 (n/a)0.45 (n/a)0.48 (n/a)0.29 (n/a)0.11 (n/a)258.40 (n/a)174.66 (n/a)152.40 (n/a)130.90 (n/a)52.15 (n/a)

test_rope[rows_32-cols_512-angle_rows_8-aie_columns_8-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.41 (-29.34%)0.37 (-13.38%)0.36 (-12.85%)0.35 (-0.45%)0.03 (-71.84%)211.10 (+0.48%)197.84 (+11.95%)204.50 (+14.76%)178.10 (+41.57%)13.75 (-60.36%)
1f9e5a9 — 2026-07-13 18:35:330.59 (n/a)0.43 (n/a)0.41 (n/a)0.35 (n/a)0.10 (n/a)210.10 (n/a)176.72 (n/a)178.20 (n/a)125.80 (n/a)34.68 (n/a)
iron/operators/softmax

test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:461.12 (+12.80%)0.87 (+16.19%)0.82 (+15.65%)0.75 (+24.57%)0.15 (-3.31%)174.80 (-19.74%)153.96 (-14.78%)159.60 (-13.54%)116.80 (-11.38%)22.19 (-31.87%)
1f9e5a9 — 2026-07-13 18:35:330.99 (n/a)0.75 (n/a)0.71 (n/a)0.60 (n/a)0.15 (n/a)217.80 (n/a)180.66 (n/a)184.60 (n/a)131.80 (n/a)32.56 (n/a)

test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.80 (-10.70%)0.66 (-9.00%)0.71 (-3.77%)0.45 (-24.25%)0.14 (+18.19%)289.40 (+32.03%)206.42 (+12.34%)184.00 (+3.90%)163.10 (+12.02%)51.73 (+73.59%)
1f9e5a9 — 2026-07-13 18:35:330.90 (n/a)0.73 (n/a)0.74 (n/a)0.60 (n/a)0.12 (n/a)219.20 (n/a)183.74 (n/a)177.10 (n/a)145.60 (n/a)29.80 (n/a)

test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.80 (+1.33%)0.65 (-2.44%)0.65 (+6.55%)0.54 (-4.44%)0.11 (+3.62%)242.70 (+4.66%)206.48 (+2.78%)203.10 (-6.15%)163.80 (-1.33%)33.79 (+11.29%)
1f9e5a9 — 2026-07-13 18:35:330.79 (n/a)0.67 (n/a)0.61 (n/a)0.57 (n/a)0.11 (n/a)231.90 (n/a)200.90 (n/a)216.40 (n/a)166.00 (n/a)30.36 (n/a)
iron/operators/swiglu_decode

test_swiglu_decode[embedding_dim_1024-hidden_dim_3584]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.00 (+2.27%)0.00 (+0.00%)0.00 (+0.00%)0.00 (-2.33%)0.00 (+191.55%)968.72 (+1.14%)950.67 (+0.52%)958.06 (+1.61%)920.15 (-2.24%)20.49 (+196.22%)
1f9e5a9 — 2026-07-13 18:35:330.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)957.84 (n/a)945.73 (n/a)942.84 (n/a)941.27 (n/a)6.92 (n/a)

test_swiglu_decode[embedding_dim_2048-hidden_dim_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.01 (+0.00%)0.01 (+0.99%)0.01 (+0.00%)0.01 (+1.32%)0.00 (-5.02%)1064.94 (-0.80%)1003.76 (-1.27%)995.30 (-0.91%)973.66 (-0.26%)37.65 (+3.83%)
1f9e5a9 — 2026-07-13 18:35:330.01 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)1073.53 (n/a)1016.68 (n/a)1004.39 (n/a)976.21 (n/a)36.26 (n/a)
iron/operators/swiglu_prefill

test_swiglu_prefill[seq_len_256-embedding_dim_2048-hidden_dim_2048-prio_accuracy_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:460.95 (-0.71%)0.94 (-0.83%)0.94 (-0.97%)0.92 (-0.07%)0.01 (-15.49%)2275.47 (+0.06%)2241.40 (+0.83%)2240.51 (+0.98%)2203.88 (+0.71%)27.18 (-15.07%)
1f9e5a9 — 2026-07-13 18:35:330.96 (n/a)0.94 (n/a)0.95 (n/a)0.92 (n/a)0.01 (n/a)2274.04 (n/a)2222.86 (n/a)2218.69 (n/a)2188.25 (n/a)32.01 (n/a)
iron/operators/transpose

test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:463.80 (+13.78%)2.71 (-12.14%)2.54 (-21.95%)2.04 (-19.25%)0.67 (+95.29%)257.40 (+23.81%)202.30 (+17.53%)206.30 (+28.14%)137.90 (-12.11%)44.07 (+105.95%)
1f9e5a9 — 2026-07-13 18:35:333.34 (n/a)3.08 (n/a)3.26 (n/a)2.52 (n/a)0.34 (n/a)207.90 (n/a)172.12 (n/a)161.00 (n/a)156.90 (n/a)21.40 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_2]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:464.64 (-22.27%)4.08 (-23.33%)3.93 (-28.67%)3.56 (-24.63%)0.52 (-5.60%)294.40 (+32.67%)260.48 (+30.96%)266.50 (+40.19%)226.00 (+28.63%)32.81 (+55.38%)
1f9e5a9 — 2026-07-13 18:35:335.97 (n/a)5.32 (n/a)5.51 (n/a)4.73 (n/a)0.56 (n/a)221.90 (n/a)198.90 (n/a)190.10 (n/a)175.70 (n/a)21.12 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
4bb8427 — 2026-06-23 23:08:223.72 (+20.32%)3.12 (+16.02%)3.04 (+6.32%)2.73 (+32.06%)0.43 (+8.15%)192.20 (-24.30%)170.32 (-14.26%)172.50 (-5.94%)141.00 (-16.86%)22.57 (-32.31%)
4d4b803 — 2026-06-22 17:54:573.09 (n/a)2.69 (n/a)2.86 (n/a)2.07 (n/a)0.40 (n/a)253.90 (n/a)198.64 (n/a)183.40 (n/a)169.60 (n/a)33.34 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8-num_batches_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:463.80 (+1.79%)2.66 (-18.33%)2.45 (-26.45%)2.15 (-23.41%)0.65 (+59.09%)243.80 (+30.58%)204.84 (+25.59%)213.80 (+36.01%)137.90 (-1.78%)39.62 (+90.37%)
1f9e5a9 — 2026-07-13 18:35:333.73 (n/a)3.26 (n/a)3.33 (n/a)2.81 (n/a)0.41 (n/a)186.70 (n/a)163.10 (n/a)157.20 (n/a)140.40 (n/a)20.81 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
4bb8427 — 2026-06-23 23:08:224.36 (+7.95%)3.26 (+13.04%)3.33 (+22.37%)2.38 (+18.97%)0.79 (-14.92%)220.70 (-15.96%)168.40 (-14.68%)157.40 (-18.28%)120.30 (-7.39%)40.78 (-34.38%)
4d4b803 — 2026-06-22 17:54:574.04 (n/a)2.89 (n/a)2.72 (n/a)2.00 (n/a)0.93 (n/a)262.60 (n/a)197.38 (n/a)192.60 (n/a)129.90 (n/a)62.14 (n/a)
Phoenix - Small

IRON

Tested on 2026_07_13_22_28_21 at commit fb80db1.

iron/operators/axpy
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_axpy[input_length_2048-num_aie_columns_1-tile_size_2048-scalar_factor_3.0]✅ 5/5303.200.04n/a
test_axpy[input_length_2048-num_aie_columns_2-tile_size_1024-scalar_factor_3.0]✅ 5/5257.800.05n/a
test_axpy[input_length_2048-num_aie_columns_4-tile_size_512-scalar_factor_3.0]✅ 5/5336.720.04n/a
iron/operators/dequant
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_dequant[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-group_size_32]✅ 5/5354.660.02n/a
test_dequant[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-group_size_32]✅ 5/5446.400.01n/a
test_dequant[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-group_size_32]✅ 5/5358.380.02n/a
test_dequant[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-group_size_32]✅ 5/5540.840.01n/a
test_dequant[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-group_size_32]✅ 5/5389.540.01n/a
test_dequant[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-group_size_32]✅ 5/5494.500.01n/a
iron/operators/elementwise_add
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_elementwise_add[input_length_2048-num_aie_columns_1-tile_size_2048]✅ 5/5315.960.04n/a
test_elementwise_add[input_length_2048-num_aie_columns_2-tile_size_1024]✅ 5/5364.800.04n/a
test_elementwise_add[input_length_2048-num_aie_columns_4-tile_size_512]✅ 5/5314.920.04n/a
iron/operators/elementwise_mul
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_elementwise_mul[input_length_2048-num_aie_columns_1-tile_size_2048]✅ 5/5283.040.04n/a
test_elementwise_mul[input_length_2048-num_aie_columns_2-tile_size_1024]✅ 5/5743.460.04n/a
test_elementwise_mul[input_length_2048-num_aie_columns_4-tile_size_512]✅ 5/5419.900.03n/a
iron/operators/gelu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_gelu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5342.160.02n/a
test_gelu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5370.560.02n/a
test_gelu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5430.000.02n/a
test_gelu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5962.780.02n/a
test_gelu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5313.560.03n/a
test_gelu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5397.180.02n/a
iron/operators/gemm
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_False-c_col_maj_False-m_48-k_96-n_16-trace_size_0-partition_N_1]✅ 5/5591.860.4017.21
test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_True-c_col_maj_True-m_48-k_96-n_16-trace_size_0-partition_N_1]✅ 5/5555.680.4117.42
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_1-b_col_maj_False-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]✅ 5/582563.340.30208.12
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_2-b_col_maj_True-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]✅ 5/524760.661.02694.09
test_gemm[M_384-K_1536-N_1792-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_32-k_48-n_64-trace_size_0-partition_N_1]✅ 5/55136.301.65431.55
test_gemm[M_64-K_512-N_256-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_16-k_64-n_64-trace_size_0-partition_N_4]✅ 5/55873.740.2211.94
iron/operators/gemv
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_gemv[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]✅ 5/5n/a0.080.08
test_gemv[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]✅ 5/5n/a3.723.72
test_gemv[M_2048-K_8192-num_aie_columns_2-tile_size_input_1-tile_size_output_1024]✅ 5/5n/a5.815.81
test_gemv[M_2048-K_8192-num_aie_columns_4-tile_size_input_1-tile_size_output_512]✅ 5/5n/a8.748.74
test_gemv[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]✅ 5/5n/a3.613.61
test_gemv[M_8192-K_2048-num_aie_columns_2-tile_size_input_4-tile_size_output_1024]✅ 5/5n/a6.736.72
test_gemv[M_8192-K_2048-num_aie_columns_4-tile_size_input_4-tile_size_output_1024]✅ 5/5n/a9.969.96
test_gemv_gelu[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]❌ 0/5n/an/an/a
test_gemv_gelu[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]❌ 0/5n/an/an/a
test_gemv_gelu[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]❌ 0/5n/an/an/a
iron/operators/layer_norm
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5461.820.02n/a
test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5319.520.03n/a
test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5419.280.02n/a
test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5352.820.02n/a
test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5374.100.02n/a
test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5768.880.02n/a
iron/operators/leaky_relu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.01]✅ 5/5399.780.02n/a
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.1]✅ 5/5317.560.03n/a
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.25]✅ 5/5393.480.02n/a
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-alpha_0.01]✅ 5/5436.260.02n/a
test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-alpha_0.01]✅ 5/5382.520.03n/a
test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-alpha_0.01]✅ 5/5394.100.02n/a
test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-alpha_0.01]✅ 5/5354.600.03n/a
test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-alpha_0.01]✅ 5/5434.520.02n/a
iron/operators/mem_copy
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_mem_copy[input_length_2048-num_cores_1-num_channels_1-bypass_False-tile_size_2048]✅ 5/5320.420.03n/a
test_mem_copy[input_length_2048-num_cores_2-num_channels_1-bypass_False-tile_size_1024]✅ 5/5435.240.02n/a
test_mem_copy[input_length_2048-num_cores_2-num_channels_2-bypass_False-tile_size_1024]✅ 5/5404.360.02n/a
test_mem_copy[input_length_2048-num_cores_4-num_channels_1-bypass_False-tile_size_512]✅ 5/5388.400.03n/a
test_mem_copy[input_length_2048-num_cores_4-num_channels_2-bypass_False-tile_size_512]✅ 5/5423.300.02n/a
test_mem_copy[input_length_2048-num_cores_8-num_channels_2-bypass_False-tile_size_256]✅ 5/5476.100.02n/a
iron/operators/relu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5366.520.03n/a
test_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5412.680.02n/a
test_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5336.840.03n/a
test_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5393.380.02n/a
test_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5527.280.02n/a
test_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5364.420.02n/a
iron/operators/rms_norm
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_False]✅ 5/5477.680.02n/a
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_True]✅ 5/5309.960.04n/a
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_False]✅ 5/5348.640.03n/a
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_True]✅ 5/5399.980.03n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_False]✅ 5/5433.540.02n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_True]✅ 5/5472.020.02n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_False]✅ 5/5395.620.02n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_True]✅ 5/5365.920.03n/a
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_False]✅ 5/5513.120.02n/a
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_True]✅ 5/5399.680.02n/a
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_False]✅ 5/5600.060.02n/a
iron/operators/rope
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_1-method_type_0]✅ 5/5328.760.32n/a
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_2-method_type_0]✅ 5/5568.560.21n/a
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_4-method_type_0]✅ 5/5420.920.25n/a
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_1-method_type_0]✅ 5/5410.400.20n/a
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_2-method_type_0]✅ 5/5592.580.15n/a
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_4-method_type_0]✅ 5/5442.780.18n/a
iron/operators/sigmoid
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_sigmoid[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5272.260.03n/a
test_sigmoid[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5303.540.03n/a
test_sigmoid[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5292.540.03n/a
test_sigmoid[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5286.180.03n/a
test_sigmoid[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5359.280.02n/a
test_sigmoid[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5440.680.02n/a
iron/operators/silu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_silu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5322.700.03n/a
test_silu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5324.440.03n/a
test_silu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5377.000.02n/a
iron/operators/softmax
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_1024]✅ 5/5503.520.29n/a
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_2048]✅ 5/5453.420.33n/a
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5493.960.29n/a
iron/operators/swiglu_decode
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_swiglu_decode[embedding_dim_1024-hidden_dim_3584]✅ 5/513848.600.00n/a
test_swiglu_decode[embedding_dim_2048-hidden_dim_2048]✅ 5/515625.580.00n/a
iron/operators/swiglu_prefill
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_swiglu_prefill[seq_len_256-embedding_dim_2048-hidden_dim_2048-prio_accuracy_False]✅ 5/520877.960.10n/a
iron/operators/tanh
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_tanh[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5338.820.03n/a
test_tanh[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5324.200.03n/a
test_tanh[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5405.060.02n/a
test_tanh[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5368.040.03n/a
test_tanh[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5433.340.03n/a
test_tanh[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5452.560.02n/a
iron/operators/transpose
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_1]✅ 5/5577.961.08n/a
test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_2]✅ 5/5659.601.78n/a
test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8-num_batches_1]✅ 5/5608.920.96n/a

Trends:

IRON Trends

iron/operators/axpy

test_axpy[input_length_2048-num_aie_columns_1-tile_size_2048-scalar_factor_3.0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.05 (+17.66%)0.04 (+48.32%)0.05 (+101.10%)0.02 (+13.70%)0.01 (+9.71%)532.20 (-12.05%)303.20 (-33.03%)244.10 (-50.26%)238.10 (-14.99%)128.33 (-15.68%)
e216c2f — 2026-07-13 21:54:420.04 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)605.10 (n/a)452.76 (n/a)490.80 (n/a)280.10 (n/a)152.20 (n/a)

test_axpy[input_length_2048-num_aie_columns_2-tile_size_1024-scalar_factor_3.0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.05 (-39.74%)0.05 (+4.20%)0.05 (+4.36%)0.04 (+76.75%)0.00 (-82.05%)298.10 (-43.41%)257.80 (-23.46%)255.10 (-4.17%)230.10 (+65.90%)26.50 (-84.12%)
e216c2f — 2026-07-13 21:54:420.09 (n/a)0.05 (n/a)0.05 (n/a)0.02 (n/a)0.03 (n/a)526.80 (n/a)336.82 (n/a)266.20 (n/a)138.70 (n/a)166.86 (n/a)

test_axpy[input_length_2048-num_aie_columns_4-tile_size_512-scalar_factor_3.0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.06 (-5.80%)0.04 (+0.59%)0.04 (-1.24%)0.02 (+2.09%)0.02 (-5.40%)651.10 (-2.05%)336.72 (-1.52%)295.80 (+1.27%)200.90 (+6.18%)180.29 (-2.91%)
e216c2f — 2026-07-13 21:54:420.06 (n/a)0.04 (n/a)0.04 (n/a)0.02 (n/a)0.02 (n/a)664.70 (n/a)341.92 (n/a)292.10 (n/a)189.20 (n/a)185.69 (n/a)
iron/operators/dequant

test_dequant[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.02 (-11.26%)0.02 (+13.32%)0.02 (+67.30%)0.01 (+1.82%)0.00 (-23.15%)503.70 (-1.79%)354.66 (-14.81%)294.50 (-40.22%)245.00 (+12.70%)110.26 (-14.30%)
e216c2f — 2026-07-13 21:54:420.02 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)512.90 (n/a)416.30 (n/a)492.60 (n/a)217.40 (n/a)128.66 (n/a)

test_dequant[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.02 (-10.13%)0.01 (+12.84%)0.01 (-10.48%)0.01 (+231.56%)0.01 (-25.70%)614.20 (-69.84%)446.40 (-41.08%)498.20 (+11.70%)237.00 (+11.27%)174.90 (-76.33%)
e216c2f — 2026-07-13 21:54:420.02 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)0.01 (n/a)2036.60 (n/a)757.66 (n/a)446.00 (n/a)213.00 (n/a)738.89 (n/a)

test_dequant[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.02 (+20.58%)0.02 (+23.58%)0.02 (+44.62%)0.01 (-5.72%)0.00 (+78.25%)514.50 (+6.06%)358.38 (-15.31%)301.00 (-30.85%)258.30 (-17.05%)111.69 (+58.27%)
e216c2f — 2026-07-13 21:54:420.02 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)485.10 (n/a)423.18 (n/a)435.30 (n/a)311.40 (n/a)70.57 (n/a)

test_dequant[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.01 (-45.43%)0.01 (-44.97%)0.01 (-50.24%)0.01 (-30.75%)0.00 (-60.86%)608.60 (+44.42%)540.84 (+78.01%)554.40 (+100.94%)447.90 (+83.27%)69.08 (+0.33%)
e216c2f — 2026-07-13 21:54:420.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.00 (n/a)421.40 (n/a)303.82 (n/a)275.90 (n/a)244.40 (n/a)68.85 (n/a)

test_dequant[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.02 (-8.69%)0.01 (+4.50%)0.02 (+41.86%)0.01 (-5.03%)0.00 (+7.66%)581.30 (+5.29%)389.54 (-2.07%)298.90 (-29.50%)282.60 (+9.53%)139.91 (+22.69%)
e216c2f — 2026-07-13 21:54:420.02 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)552.10 (n/a)397.76 (n/a)424.00 (n/a)258.00 (n/a)114.03 (n/a)

test_dequant[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.02 (+19.51%)0.01 (-0.82%)0.01 (-9.95%)0.01 (-7.71%)0.00 (+80.09%)624.40 (+8.37%)494.50 (+5.10%)518.40 (+11.03%)307.70 (-16.32%)126.51 (+62.11%)
e216c2f — 2026-07-13 21:54:420.01 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)576.20 (n/a)470.52 (n/a)466.90 (n/a)367.70 (n/a)78.04 (n/a)
iron/operators/elementwise_add

test_elementwise_add[input_length_2048-num_aie_columns_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.05 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)441.90 (n/a)315.96 (n/a)286.40 (n/a)230.10 (n/a)91.02 (n/a)

test_elementwise_add[input_length_2048-num_aie_columns_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.05 (n/a)0.04 (n/a)0.05 (n/a)0.02 (n/a)0.01 (n/a)554.30 (n/a)364.80 (n/a)263.50 (n/a)244.80 (n/a)156.32 (n/a)

test_elementwise_add[input_length_2048-num_aie_columns_4-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.05 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)488.20 (n/a)314.92 (n/a)292.50 (n/a)236.60 (n/a)99.76 (n/a)
iron/operators/elementwise_mul

test_elementwise_mul[input_length_2048-num_aie_columns_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.00 (n/a)301.80 (n/a)283.04 (n/a)287.40 (n/a)250.60 (n/a)20.22 (n/a)

test_elementwise_mul[input_length_2048-num_aie_columns_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.06 (n/a)0.04 (n/a)0.04 (n/a)0.00 (n/a)0.02 (n/a)2477.70 (n/a)743.46 (n/a)289.00 (n/a)222.80 (n/a)974.93 (n/a)

test_elementwise_mul[input_length_2048-num_aie_columns_4-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.05 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)614.80 (n/a)419.90 (n/a)428.10 (n/a)270.30 (n/a)135.75 (n/a)
iron/operators/gelu

test_gelu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (n/a)0.02 (n/a)0.03 (n/a)0.02 (n/a)0.00 (n/a)414.60 (n/a)342.16 (n/a)319.80 (n/a)275.20 (n/a)64.21 (n/a)

test_gelu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (n/a)0.02 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)557.30 (n/a)370.56 (n/a)297.60 (n/a)241.50 (n/a)138.25 (n/a)

test_gelu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)612.20 (n/a)430.00 (n/a)487.20 (n/a)224.20 (n/a)180.33 (n/a)

test_gelu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (n/a)0.02 (n/a)0.02 (n/a)0.00 (n/a)0.01 (n/a)2098.00 (n/a)962.78 (n/a)362.20 (n/a)289.60 (n/a)888.92 (n/a)

test_gelu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)411.20 (n/a)313.56 (n/a)307.10 (n/a)232.60 (n/a)80.97 (n/a)

test_gelu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)594.10 (n/a)397.18 (n/a)341.20 (n/a)291.90 (n/a)127.58 (n/a)
iron/operators/gemm

test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_False-c_col_maj_False-m_48-k_96-n_16-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:480.53 (+43.48%)0.40 (+21.80%)0.35 (-2.48%)0.26 (+21.36%)0.12 (+86.40%)851.40 (-17.60%)591.86 (-15.16%)635.10 (+2.53%)413.80 (-30.30%)182.99 (-2.69%)22.81 (+43.48%)17.21 (+21.80%)14.86 (-2.48%)11.08 (+21.36%)5.24 (+86.40%)
e216c2f — 2026-07-13 21:54:420.37 (n/a)0.33 (n/a)0.36 (n/a)0.21 (n/a)0.07 (n/a)1033.30 (n/a)697.62 (n/a)619.40 (n/a)593.70 (n/a)188.04 (n/a)15.89 (n/a)14.13 (n/a)15.24 (n/a)9.13 (n/a)2.81 (n/a)

test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_True-c_col_maj_True-m_48-k_96-n_16-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:480.47 (-7.36%)0.41 (-5.35%)0.45 (+9.08%)0.33 (-8.50%)0.07 (+20.11%)672.50 (+9.28%)555.68 (+6.79%)490.60 (-8.33%)474.10 (+7.95%)101.87 (+44.14%)19.90 (-7.36%)17.42 (-5.35%)19.24 (+9.08%)14.03 (-8.50%)3.00 (+20.11%)
e216c2f — 2026-07-13 21:54:420.50 (n/a)0.43 (n/a)0.41 (n/a)0.36 (n/a)0.06 (n/a)615.40 (n/a)520.34 (n/a)535.20 (n/a)439.20 (n/a)70.67 (n/a)21.49 (n/a)18.41 (n/a)17.63 (n/a)15.34 (n/a)2.50 (n/a)

test_gemm[M_2048-K_2048-N_2048-num_aie_columns_1-b_col_maj_False-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:480.31 (+0.79%)0.30 (-0.15%)0.30 (-0.35%)0.30 (-0.31%)0.00 (+42.02%)83993.90 (+0.31%)82563.34 (+0.16%)82705.70 (+0.35%)80751.30 (-0.78%)1202.46 (+40.98%)212.75 (+0.79%)208.12 (-0.15%)207.72 (-0.35%)204.54 (-0.31%)3.05 (+42.02%)
e216c2f — 2026-07-13 21:54:420.31 (n/a)0.31 (n/a)0.31 (n/a)0.30 (n/a)0.00 (n/a)83736.50 (n/a)82433.08 (n/a)82414.90 (n/a)81386.30 (n/a)852.92 (n/a)211.09 (n/a)208.43 (n/a)208.46 (n/a)205.17 (n/a)2.15 (n/a)

test_gemm[M_2048-K_2048-N_2048-num_aie_columns_2-b_col_maj_True-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:481.05 (+1.76%)1.02 (+0.03%)1.02 (-0.02%)0.99 (-0.57%)0.02 (+92.09%)25330.80 (+0.57%)24760.66 (-0.00%)24657.50 (+0.02%)24069.30 (-1.73%)523.99 (+90.23%)713.77 (+1.76%)694.09 (+0.03%)696.74 (-0.02%)678.22 (-0.57%)14.72 (+92.09%)
e216c2f — 2026-07-13 21:54:421.03 (n/a)1.02 (n/a)1.02 (n/a)1.00 (n/a)0.01 (n/a)25186.40 (n/a)24761.20 (n/a)24652.60 (n/a)24493.90 (n/a)275.44 (n/a)701.39 (n/a)693.89 (n/a)696.88 (n/a)682.11 (n/a)7.66 (n/a)

test_gemm[M_384-K_1536-N_1792-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_32-k_48-n_64-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:482.31 (-39.40%)1.65 (-37.25%)1.50 (-29.25%)1.24 (-22.04%)0.42 (-56.67%)6480.70 (+28.27%)5136.30 (+49.69%)5358.40 (+41.34%)3487.90 (+65.01%)1172.93 (-4.33%)606.08 (-39.40%)431.55 (-37.25%)394.51 (-29.25%)326.19 (-22.04%)111.14 (-56.67%)
e216c2f — 2026-07-13 21:54:423.81 (n/a)2.62 (n/a)2.13 (n/a)1.60 (n/a)0.98 (n/a)5052.20 (n/a)3431.40 (n/a)3791.10 (n/a)2113.70 (n/a)1225.96 (n/a)1000.11 (n/a)687.77 (n/a)557.60 (n/a)418.42 (n/a)256.51 (n/a)

test_gemm[M_64-K_512-N_256-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_16-k_64-n_64-trace_size_0-partition_N_4]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:480.31 (+5.69%)0.22 (-7.92%)0.21 (-10.17%)0.17 (-6.28%)0.06 (+23.60%)7486.30 (+6.70%)5873.74 (+10.10%)5988.40 (+11.32%)3961.50 (-5.39%)1280.69 (+19.13%)16.94 (+5.69%)11.94 (-7.92%)11.21 (-10.17%)8.96 (-6.28%)3.01 (+23.60%)
e216c2f — 2026-07-13 21:54:420.30 (n/a)0.24 (n/a)0.23 (n/a)0.18 (n/a)0.05 (n/a)7016.30 (n/a)5334.86 (n/a)5379.50 (n/a)4187.00 (n/a)1075.07 (n/a)16.03 (n/a)12.97 (n/a)12.47 (n/a)9.56 (n/a)2.43 (n/a)
iron/operators/gemv

test_gemv[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:480.13 (n/a)0.08 (n/a)0.07 (n/a)0.02 (n/a)0.05 (n/a)0.12 (n/a)0.08 (n/a)0.07 (n/a)0.02 (n/a)0.04 (n/a)

test_gemv[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:483.94 (n/a)3.72 (n/a)3.74 (n/a)3.44 (n/a)0.18 (n/a)3.94 (n/a)3.72 (n/a)3.74 (n/a)3.44 (n/a)0.18 (n/a)

test_gemv[M_2048-K_8192-num_aie_columns_2-tile_size_input_1-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:486.07 (-18.83%)5.81 (-8.27%)5.65 (-9.04%)5.63 (-0.83%)0.23 (-69.18%)6.07 (-18.83%)5.81 (-8.27%)5.65 (-9.04%)5.63 (-0.83%)0.23 (-69.18%)
e216c2f — 2026-07-13 21:54:427.48 (n/a)6.33 (n/a)6.21 (n/a)5.68 (n/a)0.73 (n/a)7.48 (n/a)6.33 (n/a)6.21 (n/a)5.68 (n/a)0.73 (n/a)

test_gemv[M_2048-K_8192-num_aie_columns_4-tile_size_input_1-tile_size_output_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4810.92 (-21.20%)8.74 (-18.40%)8.48 (-19.62%)6.90 (-8.75%)1.48 (-49.39%)10.92 (-21.20%)8.74 (-18.40%)8.47 (-19.62%)6.90 (-8.75%)1.48 (-49.39%)
e216c2f — 2026-07-13 21:54:4213.86 (n/a)10.71 (n/a)10.55 (n/a)7.56 (n/a)2.93 (n/a)13.85 (n/a)10.71 (n/a)10.54 (n/a)7.56 (n/a)2.93 (n/a)

test_gemv[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:483.78 (n/a)3.61 (n/a)3.61 (n/a)3.46 (n/a)0.14 (n/a)3.78 (n/a)3.61 (n/a)3.61 (n/a)3.46 (n/a)0.14 (n/a)

test_gemv[M_8192-K_2048-num_aie_columns_2-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:487.55 (+0.89%)6.73 (+1.36%)6.96 (+2.76%)5.79 (+1.26%)0.82 (+21.89%)7.54 (+0.89%)6.72 (+1.36%)6.96 (+2.76%)5.79 (+1.26%)0.82 (+21.89%)
e216c2f — 2026-07-13 21:54:427.48 (n/a)6.64 (n/a)6.77 (n/a)5.72 (n/a)0.67 (n/a)7.48 (n/a)6.63 (n/a)6.77 (n/a)5.71 (n/a)0.67 (n/a)

test_gemv[M_8192-K_2048-num_aie_columns_4-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
fb80db1 — 2026-07-13 22:24:4813.82 (+23.57%)9.96 (+1.17%)8.37 (-21.38%)7.03 (-12.75%)3.00 (+112.02%)13.81 (+23.57%)9.96 (+1.17%)8.37 (-21.38%)7.03 (-12.75%)3.00 (+112.02%)
e216c2f — 2026-07-13 21:54:4211.18 (n/a)9.85 (n/a)10.65 (n/a)8.06 (n/a)1.41 (n/a)11.18 (n/a)9.84 (n/a)10.64 (n/a)8.06 (n/a)1.41 (n/a)

test_gemv_batched[M_1024-K_1024-num_aie_columns_1-tile_size_input_1-tile_size_output_64-num_batches_2]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
e216c2f — 2026-07-13 21:54:423.15 (n/a)2.42 (n/a)2.82 (n/a)1.13 (n/a)0.83 (n/a)3.14 (n/a)2.42 (n/a)2.82 (n/a)1.13 (n/a)0.83 (n/a)

test_gemv_batched[M_1026-K_64-num_aie_columns_1-tile_size_input_1-tile_size_output_2-num_batches_2]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
e216c2f — 2026-07-13 21:54:420.47 (n/a)0.33 (n/a)0.35 (n/a)0.13 (n/a)0.14 (n/a)0.46 (n/a)0.32 (n/a)0.34 (n/a)0.12 (n/a)0.14 (n/a)

test_gemv_batched[M_256-K_128-num_aie_columns_1-tile_size_input_1-tile_size_output_256-num_batches_4]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
e216c2f — 2026-07-13 21:54:420.63 (n/a)0.50 (n/a)0.58 (n/a)0.08 (n/a)0.24 (n/a)0.62 (n/a)0.49 (n/a)0.58 (n/a)0.07 (n/a)0.24 (n/a)

test_gemv_batched[M_64-K_1536-num_aie_columns_1-tile_size_input_1-tile_size_output_64-num_batches_8]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
e216c2f — 2026-07-13 21:54:422.27 (n/a)1.72 (n/a)1.60 (n/a)1.11 (n/a)0.45 (n/a)2.23 (n/a)1.69 (n/a)1.57 (n/a)1.10 (n/a)0.44 (n/a)

test_gemv_gelu[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]

No metrics available.

test_gemv_gelu[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]

No metrics available.

test_gemv_gelu[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]

No metrics available.

iron/operators/layer_norm

test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.00 (n/a)567.40 (n/a)461.82 (n/a)486.40 (n/a)341.90 (n/a)97.84 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)482.60 (n/a)319.52 (n/a)262.70 (n/a)237.70 (n/a)106.05 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)520.20 (n/a)419.28 (n/a)426.60 (n/a)273.50 (n/a)105.44 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (n/a)0.02 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)473.00 (n/a)352.82 (n/a)302.10 (n/a)286.40 (n/a)82.40 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)492.00 (n/a)374.10 (n/a)404.60 (n/a)232.50 (n/a)108.12 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (n/a)0.02 (n/a)0.02 (n/a)0.00 (n/a)0.01 (n/a)1904.30 (n/a)768.88 (n/a)543.20 (n/a)297.90 (n/a)646.92 (n/a)
iron/operators/leaky_relu

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (-11.68%)0.02 (-18.15%)0.02 (-24.52%)0.01 (+2.44%)0.01 (-18.93%)562.70 (-2.38%)399.78 (+18.93%)356.50 (+32.48%)279.00 (+13.23%)120.58 (-12.39%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.03 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)576.40 (n/a)336.14 (n/a)269.10 (n/a)246.40 (n/a)137.63 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (+6.53%)0.03 (+15.84%)0.03 (+34.28%)0.02 (-11.53%)0.01 (+33.68%)490.60 (+13.04%)317.56 (-11.09%)278.60 (-25.53%)246.30 (-6.14%)98.35 (+54.47%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.00 (n/a)434.00 (n/a)357.16 (n/a)374.10 (n/a)262.40 (n/a)63.67 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.25]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (-7.89%)0.02 (+1.58%)0.02 (+42.87%)0.01 (-9.57%)0.01 (-17.21%)602.70 (+10.59%)393.48 (-3.71%)336.30 (-30.00%)261.60 (+8.55%)147.11 (-0.14%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)545.00 (n/a)408.62 (n/a)480.40 (n/a)241.00 (n/a)147.32 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (-4.74%)0.02 (+9.29%)0.02 (-1.20%)0.01 (+245.07%)0.01 (-26.47%)578.80 (-71.02%)436.26 (-38.98%)487.80 (+1.22%)255.70 (+4.97%)152.92 (-78.98%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.02 (n/a)0.00 (n/a)0.01 (n/a)1997.20 (n/a)714.94 (n/a)481.90 (n/a)243.60 (n/a)727.43 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.05 (+75.15%)0.03 (+25.12%)0.02 (-3.45%)0.01 (-9.21%)0.01 (+202.66%)578.80 (+10.16%)382.52 (-8.19%)420.80 (+3.59%)174.00 (-42.91%)156.09 (+83.60%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.00 (n/a)525.40 (n/a)416.66 (n/a)406.20 (n/a)304.80 (n/a)85.01 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (+25.13%)0.02 (+35.53%)0.02 (+27.72%)0.01 (+139.67%)0.01 (+10.03%)546.40 (-58.27%)394.10 (-37.05%)421.70 (-21.70%)183.20 (-20.07%)137.39 (-66.05%)
e216c2f — 2026-07-13 21:54:420.04 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)1309.50 (n/a)626.08 (n/a)538.60 (n/a)229.20 (n/a)404.72 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (+104.36%)0.03 (+65.67%)0.03 (+78.66%)0.01 (+9.66%)0.01 (+309.93%)584.60 (-8.81%)354.60 (-30.75%)275.60 (-44.02%)198.60 (-51.06%)163.71 (+86.99%)
e216c2f — 2026-07-13 21:54:420.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.00 (n/a)641.10 (n/a)512.08 (n/a)492.30 (n/a)405.80 (n/a)87.55 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (+8.13%)0.02 (+3.14%)0.02 (+7.21%)0.01 (-23.20%)0.01 (+52.61%)697.40 (+30.18%)434.52 (+2.57%)363.10 (-6.73%)307.40 (-7.52%)161.48 (+80.88%)
e216c2f — 2026-07-13 21:54:420.02 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.00 (n/a)535.70 (n/a)423.62 (n/a)389.30 (n/a)332.40 (n/a)89.28 (n/a)
iron/operators/mem_copy

test_mem_copy[input_length_2048-num_cores_1-num_channels_1-bypass_False-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (+5.37%)0.03 (+17.37%)0.02 (+7.31%)0.02 (+25.74%)0.01 (-0.27%)401.20 (-20.46%)320.42 (-16.51%)335.40 (-6.81%)226.70 (-5.07%)83.13 (-27.85%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)504.40 (n/a)383.80 (n/a)359.90 (n/a)238.80 (n/a)115.21 (n/a)

test_mem_copy[input_length_2048-num_cores_2-num_channels_1-bypass_False-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (-25.83%)0.02 (-17.57%)0.02 (-14.33%)0.01 (-15.60%)0.01 (-18.91%)580.90 (+18.48%)435.24 (+21.89%)437.80 (+16.72%)294.30 (+34.81%)139.30 (+28.73%)
e216c2f — 2026-07-13 21:54:420.04 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)490.30 (n/a)357.08 (n/a)375.10 (n/a)218.30 (n/a)108.21 (n/a)

test_mem_copy[input_length_2048-num_cores_2-num_channels_2-bypass_False-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (-2.74%)0.02 (-4.87%)0.02 (-5.02%)0.02 (-6.51%)0.01 (+12.46%)513.30 (+6.96%)404.36 (+7.54%)440.30 (+5.28%)264.00 (+2.80%)117.99 (+27.22%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)479.90 (n/a)376.00 (n/a)418.20 (n/a)256.80 (n/a)92.75 (n/a)

test_mem_copy[input_length_2048-num_cores_4-num_channels_1-bypass_False-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (-5.81%)0.03 (-7.47%)0.03 (+10.25%)0.01 (-20.05%)0.01 (+2.86%)643.80 (+25.08%)388.40 (+14.98%)261.70 (-9.29%)212.40 (+6.15%)206.31 (+40.84%)
e216c2f — 2026-07-13 21:54:420.04 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)514.70 (n/a)337.80 (n/a)288.50 (n/a)200.10 (n/a)146.49 (n/a)

test_mem_copy[input_length_2048-num_cores_4-num_channels_2-bypass_False-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (-2.93%)0.02 (-4.31%)0.02 (+22.96%)0.01 (-16.24%)0.01 (-6.01%)678.80 (+19.38%)423.30 (+5.71%)360.50 (-18.66%)240.80 (+3.04%)173.09 (+21.21%)
e216c2f — 2026-07-13 21:54:420.04 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)568.60 (n/a)400.42 (n/a)443.20 (n/a)233.70 (n/a)142.81 (n/a)

test_mem_copy[input_length_2048-num_cores_8-num_channels_2-bypass_False-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (+74.75%)0.02 (+23.23%)0.02 (+5.71%)0.01 (+6.54%)0.01 (+191.42%)608.60 (-6.14%)476.10 (-8.33%)510.60 (-5.39%)205.50 (-42.77%)162.70 (+56.35%)
e216c2f — 2026-07-13 21:54:420.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.00 (n/a)648.40 (n/a)519.36 (n/a)539.70 (n/a)359.10 (n/a)104.07 (n/a)
iron/operators/rms_norm

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (-8.12%)0.02 (-32.81%)0.02 (-46.70%)0.01 (-52.83%)0.01 (+81.83%)783.90 (+111.98%)477.68 (+70.14%)488.70 (+87.60%)256.50 (+8.87%)209.60 (+296.25%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.00 (n/a)369.80 (n/a)280.76 (n/a)260.50 (n/a)235.60 (n/a)52.90 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.05 (+8.17%)0.04 (+44.53%)0.04 (+93.96%)0.02 (+18.11%)0.01 (-10.22%)506.00 (-15.33%)309.96 (-33.50%)279.10 (-48.44%)232.60 (-7.55%)111.59 (-28.56%)
e216c2f — 2026-07-13 21:54:420.05 (n/a)0.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)597.60 (n/a)466.14 (n/a)541.30 (n/a)251.60 (n/a)156.20 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (-11.93%)0.03 (+15.07%)0.03 (+64.91%)0.01 (-4.50%)0.01 (-29.01%)573.90 (+4.71%)348.64 (-16.76%)299.30 (-39.36%)281.40 (+13.56%)126.17 (-13.35%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)548.10 (n/a)418.86 (n/a)493.60 (n/a)247.80 (n/a)145.61 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (-35.52%)0.03 (-17.54%)0.03 (-6.67%)0.01 (-20.58%)0.01 (-40.19%)688.80 (+25.92%)399.98 (+16.09%)320.50 (+7.12%)229.80 (+55.06%)186.14 (+21.88%)
e216c2f — 2026-07-13 21:54:420.07 (n/a)0.04 (n/a)0.03 (n/a)0.02 (n/a)0.02 (n/a)547.00 (n/a)344.54 (n/a)299.20 (n/a)148.20 (n/a)152.73 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (-24.25%)0.02 (-17.63%)0.02 (-21.45%)0.01 (+3.03%)0.01 (-32.06%)666.40 (-2.94%)433.54 (+14.38%)362.20 (+27.27%)310.40 (+32.03%)152.65 (-17.19%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.03 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)686.60 (n/a)379.02 (n/a)284.60 (n/a)235.10 (n/a)184.33 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (-18.53%)0.02 (-20.99%)0.02 (-33.31%)0.01 (-14.16%)0.01 (-12.12%)684.00 (+16.49%)472.02 (+27.61%)484.90 (+49.94%)287.50 (+22.76%)177.07 (+20.95%)
e216c2f — 2026-07-13 21:54:420.04 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)587.20 (n/a)369.90 (n/a)323.40 (n/a)234.20 (n/a)146.40 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.04 (+12.65%)0.02 (-2.38%)0.02 (-12.88%)0.01 (-5.41%)0.01 (+45.12%)601.20 (+5.73%)395.62 (+8.27%)376.10 (+14.77%)230.30 (-11.22%)153.28 (+28.45%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)568.60 (n/a)365.40 (n/a)327.70 (n/a)259.40 (n/a)119.32 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (+32.25%)0.03 (+36.02%)0.03 (+58.30%)0.02 (+14.99%)0.01 (+82.60%)560.60 (-13.03%)365.92 (-23.51%)293.70 (-36.83%)270.50 (-24.40%)122.13 (+17.36%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.00 (n/a)644.60 (n/a)478.40 (n/a)464.90 (n/a)357.80 (n/a)104.07 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (+6.53%)0.02 (-6.71%)0.02 (-12.97%)0.01 (-56.52%)0.01 (+80.81%)1093.00 (+129.96%)513.12 (+36.45%)488.00 (+14.90%)243.30 (-6.13%)346.23 (+270.20%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)475.30 (n/a)376.04 (n/a)424.70 (n/a)259.20 (n/a)93.53 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (-20.05%)0.02 (-13.33%)0.02 (-15.38%)0.02 (+26.85%)0.01 (-35.58%)530.90 (-21.17%)399.68 (+5.75%)374.30 (+18.19%)281.10 (+25.10%)120.58 (-34.62%)
e216c2f — 2026-07-13 21:54:420.04 (n/a)0.03 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)673.50 (n/a)377.96 (n/a)316.70 (n/a)224.70 (n/a)184.43 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.03 (+26.89%)0.02 (+4.71%)0.02 (+3.54%)0.01 (-6.70%)0.01 (+50.04%)1065.70 (+7.18%)600.06 (+3.71%)526.90 (-3.43%)255.30 (-21.18%)297.96 (+19.34%)
e216c2f — 2026-07-13 21:54:420.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)994.30 (n/a)578.58 (n/a)545.60 (n/a)323.90 (n/a)249.69 (n/a)
iron/operators/rope

test_rope[rows_32-cols_512-angle_rows_32-aie_columns_1-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.40 (+35.08%)0.32 (+31.53%)0.34 (+41.43%)0.20 (+9.59%)0.08 (+61.22%)495.00 (-8.76%)328.76 (-21.99%)290.80 (-29.30%)243.60 (-25.98%)97.94 (+14.48%)
e216c2f — 2026-07-13 21:54:420.30 (n/a)0.24 (n/a)0.24 (n/a)0.18 (n/a)0.05 (n/a)542.50 (n/a)421.44 (n/a)411.30 (n/a)329.10 (n/a)85.56 (n/a)

test_rope[rows_32-cols_512-angle_rows_32-aie_columns_2-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.37 (+6.03%)0.21 (-20.49%)0.20 (-34.29%)0.09 (-43.84%)0.10 (+8.76%)1086.40 (+78.07%)568.56 (+36.67%)486.40 (+52.19%)262.40 (-5.68%)307.77 (+85.01%)
e216c2f — 2026-07-13 21:54:420.35 (n/a)0.27 (n/a)0.31 (n/a)0.16 (n/a)0.09 (n/a)610.10 (n/a)416.02 (n/a)319.60 (n/a)278.20 (n/a)166.35 (n/a)

test_rope[rows_32-cols_512-angle_rows_32-aie_columns_4-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.37 (-23.53%)0.25 (-5.75%)0.23 (+9.74%)0.18 (+14.84%)0.08 (-39.12%)552.70 (-12.93%)420.92 (-2.41%)433.50 (-8.87%)266.10 (+30.76%)121.93 (-27.56%)
e216c2f — 2026-07-13 21:54:420.48 (n/a)0.27 (n/a)0.21 (n/a)0.15 (n/a)0.13 (n/a)634.80 (n/a)431.30 (n/a)475.70 (n/a)203.50 (n/a)168.32 (n/a)

test_rope[rows_32-cols_512-angle_rows_8-aie_columns_1-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.29 (+23.06%)0.20 (+21.22%)0.22 (+44.78%)0.12 (+6.68%)0.07 (+48.86%)592.80 (-6.26%)410.40 (-13.72%)339.70 (-30.93%)258.40 (-18.72%)149.77 (+20.48%)
e216c2f — 2026-07-13 21:54:420.23 (n/a)0.16 (n/a)0.15 (n/a)0.12 (n/a)0.05 (n/a)632.40 (n/a)475.66 (n/a)491.80 (n/a)317.90 (n/a)124.31 (n/a)

test_rope[rows_32-cols_512-angle_rows_8-aie_columns_2-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.25 (+30.93%)0.15 (-6.94%)0.14 (-10.22%)0.07 (-46.99%)0.07 (+119.66%)1102.30 (+88.65%)592.58 (+24.56%)538.30 (+11.38%)295.70 (-23.61%)302.34 (+242.03%)
e216c2f — 2026-07-13 21:54:420.19 (n/a)0.16 (n/a)0.15 (n/a)0.13 (n/a)0.03 (n/a)584.30 (n/a)475.72 (n/a)483.30 (n/a)387.10 (n/a)88.40 (n/a)

test_rope[rows_32-cols_512-angle_rows_8-aie_columns_4-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.25 (-7.70%)0.18 (-7.35%)0.16 (+13.65%)0.13 (+2.35%)0.05 (-33.53%)553.30 (-2.30%)442.78 (+2.16%)456.30 (-12.01%)293.60 (+8.34%)105.12 (-27.68%)
e216c2f — 2026-07-13 21:54:420.27 (n/a)0.19 (n/a)0.14 (n/a)0.13 (n/a)0.07 (n/a)566.30 (n/a)433.40 (n/a)518.60 (n/a)271.00 (n/a)145.36 (n/a)
iron/operators/softmax

test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.50 (+13.57%)0.29 (+6.63%)0.22 (-20.03%)0.22 (+301.34%)0.12 (-12.75%)606.20 (-75.08%)503.52 (-38.83%)587.90 (+25.06%)261.10 (-11.94%)146.67 (-83.76%)
e216c2f — 2026-07-13 21:54:420.44 (n/a)0.27 (n/a)0.28 (n/a)0.05 (n/a)0.14 (n/a)2432.90 (n/a)823.14 (n/a)470.10 (n/a)296.50 (n/a)903.28 (n/a)

test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.48 (-6.91%)0.33 (-8.20%)0.36 (+4.85%)0.19 (+1.18%)0.12 (-8.76%)684.30 (-1.16%)453.42 (+8.40%)360.00 (-4.61%)275.30 (+7.41%)182.14 (+2.90%)
e216c2f — 2026-07-13 21:54:420.51 (n/a)0.36 (n/a)0.35 (n/a)0.19 (n/a)0.13 (n/a)692.30 (n/a)418.28 (n/a)377.40 (n/a)256.30 (n/a)177.01 (n/a)

test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.47 (+8.74%)0.29 (-7.19%)0.24 (-6.96%)0.21 (-1.34%)0.11 (+5.85%)628.00 (+1.37%)493.96 (+8.08%)550.70 (+7.47%)278.90 (-8.04%)136.48 (-0.69%)
e216c2f — 2026-07-13 21:54:420.43 (n/a)0.31 (n/a)0.26 (n/a)0.21 (n/a)0.10 (n/a)619.50 (n/a)457.04 (n/a)512.40 (n/a)303.30 (n/a)137.43 (n/a)
iron/operators/swiglu_decode

test_swiglu_decode[embedding_dim_1024-hidden_dim_3584]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.00 (+16.67%)0.00 (+28.57%)0.00 (+50.00%)0.00 (+0.00%)0.00 (+15.92%)23027.56 (+0.56%)13848.60 (-24.97%)13850.24 (-37.84%)5848.61 (-13.06%)6592.85 (-3.90%)
e216c2f — 2026-07-13 21:54:420.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)22899.76 (n/a)18458.31 (n/a)22279.89 (n/a)6727.16 (n/a)6860.73 (n/a)

test_swiglu_decode[embedding_dim_2048-hidden_dim_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.00 (-21.43%)0.00 (-43.40%)0.00 (-58.33%)0.00 (+0.00%)0.00 (-26.48%)19228.55 (-15.73%)15625.58 (+55.73%)17459.18 (+153.64%)7248.68 (+19.69%)4781.84 (-33.30%)
e216c2f — 2026-07-13 21:54:420.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)22818.84 (n/a)10033.62 (n/a)6883.34 (n/a)6056.00 (n/a)7169.71 (n/a)
iron/operators/swiglu_prefill

test_swiglu_prefill[seq_len_256-embedding_dim_2048-hidden_dim_2048-prio_accuracy_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:480.14 (-6.50%)0.10 (-3.59%)0.09 (-4.28%)0.08 (+9.33%)0.03 (-18.17%)25217.24 (-8.55%)20877.96 (+1.78%)22332.66 (+4.51%)14722.53 (+6.94%)4461.98 (-18.23%)
e216c2f — 2026-07-13 21:54:420.15 (n/a)0.11 (n/a)0.10 (n/a)0.08 (n/a)0.03 (n/a)27574.44 (n/a)20512.79 (n/a)21368.30 (n/a)13767.47 (n/a)5456.68 (n/a)
iron/operators/transpose

test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:482.10 (+93.87%)1.08 (+26.93%)0.83 (+5.21%)0.65 (+3.80%)0.59 (+223.96%)811.10 (-3.67%)577.96 (-9.90%)633.70 (-4.95%)249.60 (-48.41%)217.45 (+54.90%)
e216c2f — 2026-07-13 21:54:421.08 (n/a)0.85 (n/a)0.79 (n/a)0.62 (n/a)0.18 (n/a)842.00 (n/a)641.50 (n/a)666.70 (n/a)483.80 (n/a)140.38 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_2]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:482.58 (+4.97%)1.78 (+1.07%)1.52 (-32.14%)1.06 (+236.47%)0.67 (-27.66%)990.80 (-70.28%)659.60 (-39.22%)691.30 (+47.37%)406.10 (-4.74%)243.51 (-80.75%)
e216c2f — 2026-07-13 21:54:422.46 (n/a)1.76 (n/a)2.24 (n/a)0.31 (n/a)0.93 (n/a)3333.70 (n/a)1085.28 (n/a)469.10 (n/a)426.30 (n/a)1264.95 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
4bb8427 — 2026-06-25 20:01:371.58 (-2.67%)1.21 (+5.65%)1.29 (+41.89%)0.66 (-20.05%)0.38 (-2.53%)792.90 (+25.08%)479.12 (-3.80%)406.80 (-29.52%)331.10 (+2.76%)189.18 (+26.62%)
4bb8427 — 2026-06-23 22:46:491.63 (n/a)1.14 (n/a)0.91 (n/a)0.83 (n/a)0.38 (n/a)633.90 (n/a)498.02 (n/a)577.20 (n/a)322.20 (n/a)149.40 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8-num_batches_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
fb80db1 — 2026-07-13 22:24:481.63 (+7.98%)0.96 (-10.88%)0.80 (-15.97%)0.63 (-0.44%)0.40 (+11.66%)836.50 (+0.44%)608.92 (+13.90%)658.80 (+19.00%)321.80 (-7.40%)201.20 (+4.01%)
e216c2f — 2026-07-13 21:54:421.51 (n/a)1.08 (n/a)0.95 (n/a)0.63 (n/a)0.36 (n/a)832.80 (n/a)534.60 (n/a)553.60 (n/a)347.50 (n/a)193.43 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
4bb8427 — 2026-06-25 20:01:371.75 (-15.44%)1.32 (-0.50%)1.36 (+22.47%)0.77 (-19.20%)0.36 (-21.75%)678.10 (+23.76%)430.02 (-0.09%)385.10 (-18.34%)300.10 (+18.29%)146.96 (+19.46%)
4bb8427 — 2026-06-23 22:46:492.07 (n/a)1.32 (n/a)1.11 (n/a)0.96 (n/a)0.46 (n/a)547.90 (n/a)430.40 (n/a)471.60 (n/a)253.70 (n/a)123.02 (n/a)
Phoenix - Examples

IRON

Tested on 2026_07_13_22_33_35 at commit fb80db1.

Trends:

IRON Trends

@atassis

atassis commented Jul 13, 2026

Copy link
Copy Markdown
Contributor Author

I am already sticking to a generic "run an arbitrary Kernel in-place on the output" epilogue slot, not a one-off gelu branch.

The genericity matters to me because I'm building a full NPU engine backend on XDNA2, so composing generic primitives instead of hand-fusing per model is the design rule I hold to - a generic epilogue slot fits that squarely. The harder piece I'm on right now is the resident fused dataflow: keeping the whole transformer block on-chip across dispatches, with on-chip KV and decode attention, to cut out the host round-trips - and that only holds together if the primitives stay composable. If you have a take on whether that's the right direction to push, I'd genuinely value it; you all have far more context on where the toolchain is heading.

On the PR itself: I'll rebase onto the latest devel (the conflict is just from my batched-gemv PR that merged) and clear the lint and any other CI so it lands clean. Happy to do the generic epilogue as a follow-on once this lands.

Adds an opt-in `epilogue` parameter to GEMV. `epilogue="none"` (default) leaves
the output unchanged and is byte-identical to before; `epilogue="gelu"` applies a
GELU (tanh approximation) to each output tile inside the producing core, fusing an
activation that would otherwise be a separate elementwise pass over the GEMV output.

The GELU is applied once per full output tile in core_body (after the matvec
inner-loop fills all rows), not per matvec call, whose m_input tile can be smaller
than the 16-wide activation vector.

To avoid duplicating the GELU math, aie2p/gelu.cc is refactored: the per-16-lane
computation is factored into a shared helper used by both the existing out-of-place
gelu_tanh_approx_bf16 and a new in-place gelu_tile_bf16. In-place is aliasing-correct
(a single restrict pointer, each slot read then written) whereas reusing the
out-of-place kernel with input==output would violate its restrict contract. Because
gelu.cc is aie2p-only, the fused epilogue is NPU2-only; GEMV raises NotImplementedError
if requested on NPU1. When enabled the core links a (matvec, gelu) archive.

Test: test_gemv_gelu compares the fused output against a gelu(A @ B) golden across
three shapes, with the standard latency/bandwidth/throughput metrics. Verified on
NPU2 (15/15 pass).
@atassis atassis force-pushed the feat/gemv-gelu-epilogue branch from a33ac61 to 93d59b5 Compare July 13, 2026 22:43
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants