NVIDIA Sol-Attn for ComfyUI / Triton kernel on SM89 - SM121, with zero-copy MiniMax H3 nodes: memory-efficient attention, scheduled tau with graph preview, and feed-forward chunking. Measured 1.14–1.44× vs SageAttention and −37% MLP peak VRAM on H3
triton attention video-generation memory-optimization diffusion-models blackwell memory-op sparse-attention comfyui sageattention comfyui-custom-nodes trito minimax-h3 sol-attn
-
Updated
Aug 13, 2026 - Python