ComfyUI Node
AutoMultiGPU - DisTorch Engine (Layer Sharding)
Distributes UNet/DiT layers 50/50 across Dual GPUs for massive models with zero OOM across all architectures.
AutoMultiGPU - DisTorch Engine (Layer Sharding)
- MODEL
- CLIP
- VAE
◄unet_name▾►
◄clip_name1none►
◄clip_name2none►
◄vae_namepixel_space►
◄model_typeauto►
◄sharding_ratio50% GPU0 / 50% GPU1►
◄weight_dtypedefault►
◄force_distorch_patchtrue►
CategoryAutoMultiGPU/Engines
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| unet_name | COMBO | Select UNet / DiT model (.safetensors or .gguf) | |
| clip_name1 | COMBO | none | First Text Encoder (.safetensors or .gguf) |
| clip_name2 | COMBO | none | Optional Second Text Encoder |
| vae_name | COMBO | pixel_space | VAE model |
| model_type | COMBO | auto | Architecture model type |
| sharding_ratio | COMBO | 50% GPU0 / 50% GPU1 | 3 options: 50% GPU0 / 50% GPU1, 60% GPU0 / 40% GPU1, 40% GPU0 / 60% GPU1 |
| weight_dtypeopt | COMBO | default | 7 options: default, fp8_e4m3fn, fp8_e4m3fn_fast, fp8_e5m2, float16, bfloat16, +1 |
| force_distorch_patchopt | BOOLEAN | true | Enforce DisTorch2 memory patcher for zero-OOM sharding |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |
| CLIP | CLIP | — |
| VAE | VAE | — |