ComfyUI Node
Auto Checkpoint Loader (MultiGPU)
Pure PyTorch native checkpoint loader with automatic multi-GPU distribution.
Auto Checkpoint Loader (MultiGPU)
- MODEL
- CLIP
- VAE
◄ckpt_name▾►
◄distribution_modeAuto: UNet GPU0 / Encoders GPU1►
◄manual_unet_devicecpu►
◄manual_clip_devicecpu►
◄manual_vae_devicecpu►
CategoryAutoMultiGPU
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| ckpt_name | COMBO | The checkpoint to load. | |
| distribution_mode | COMBO | Auto: UNet GPU0 / Encoders GPU1 | 4 options: Auto: UNet GPU0 / Encoders GPU1, Auto: Balanced Free VRAM, Auto: Split Video (GPU0 Sampler / GPU1 VAE), Manual Override |
| manual_unet_deviceopt | COMBO | cpu | 1 options: cpu |
| manual_clip_deviceopt | COMBO | cpu | 1 options: cpu |
| manual_vae_deviceopt | COMBO | cpu | 1 options: cpu |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | Model (UNet/DiT) routed to Primary/Compute GPU. |
| CLIP | CLIP | CLIP (Text Encoder) routed to Secondary GPU. |
| VAE | VAE | VAE (Encode/Decode) routed to Secondary GPU. |