You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
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
Progressive one-pass MiniMax H3 video sampling for ComfyUI, jointly using Sol-Attn and combining cross-resolution noise coupling and latent upscaling to address performance bottlenecks and improve generation speed when generating high-resolution, long-form videos with MiniMax H3.
NVIDIA's Sol-Attn training-free sparse attention as an opt-in ComfyUI model patch - including consumer Blackwell (SM120 / RTX 50-series), which upstream gates off.