Skip to content

💻 Hardware Requirements

MiniMax H3 model files are large and rely on ComfyUI offloading. Resolution, duration and concurrent model loading can increase both VRAM and system RAM requirements.

Tested configurations

Provisioning GPU Model Purpose Pod RAM Tested output
NVIDIA LVRAM RTX 3090/4090 24 GB Pruned INT8 ConvRot Lowest cost and maximum compatibility 50 GB 0.9 MP, 15 seconds
NVIDIA HVRAM L40S 48 GB Full INT8 ConvRot Quality and longer video 80 GB 0.9 MP, 20 seconds, 24 fps
Blackwell LVRAM RTX 5090 32 GB Pruned INT8 ConvRot Compatible low-VRAM profile for Blackwell 70 GB 1.0 MP, 15 seconds, 24 fps
Blackwell HVRAM RTX PRO 6000 96 GB Full MXFP8 (FP8 scaled) Maximum quality and speed 70 GB 2 MP, 15 seconds, 24 fps
  • Generation limits depend on resolution, duration, model selection and offloading. Allocate additional system RAM for larger workloads.