GPU: NVIDIA GB10 (ASUS GX10), 128 GB unified memory (ComfyUI: Total VRAM 124544 MB, total RAM 124544 MB)
Driver: 580.178.04, CUDA 13.0
Python 3.12.3, torch 2.14.1+cu130
ComfyUI: master fd7639f477050e5c25b68fc38a9c2a6b6d988211 (the file reports version 0.43.0)
comfy-aimdo 0.6.0 (dynamic VRAM), comfy-kitchen 0.2.40
Attention: ComfyUI log line "Using pytorch attention" (dense; no sparse or SoL attention node in the graph)
Memory before the first job: 4,495 MiB used; swap of 1,521 MiB was already in use before the run
Model card: https://huggingface.co/Efficient-Large-Model/Open-H3-Regenerate-2K (revision 5809c33730c18b6b1c827b40974855a4288641b3)
NVIDIA code README: https://github.com/NVlabs/Sana/tree/sol-engine/models/Open-H3-Regenerate-2K (781e8341fb3b29d3aea5d50265772e9619340b66)
ComfyUI conversion by father558: https://huggingface.co/father558/H3-Open-Regenerate-2K-ComfyUI-r128 (df8d07f9715f2da78dc1d19bbdb6714a5242668f)
SwiGLU-fixed LoRA by DrakenZA: https://huggingface.co/DrakenZA/Open-H3-Regenerate-2K-ComfyUI (f81f29e4daed980eaf70ca3c653257911f780d6d)
Comfy-Org weights: https://huggingface.co/Comfy-Org/MiniMax-H3 (912b6e38b2092ec4d2e86d24183e0d59ea059d48)