ComfyUI Node
VRAM Budget Gate (Shared GPU)
Compares device-wide used VRAM plus an estimate for the next step against a budget. On a breach it moves a torch.nn.Module to CPU, or raises if hard_stop is on; any other MODEL (such as a ModelPatcher) is passed through and the breach is only reported.
VRAM Budget Gate (Shared GPU)
- model
- latent
- model
- latent
- vram_usage_gb
◄max_vram_gb80.0►
◄run_idrun►
◄hard_stopfalse►
◄probe_gb-1.0►
CategoryParametric/System
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | The model to use. | |
| latent | LATENT | The latent to work on. | |
| max_vram_gb | FLOAT | 80.01–512 | Max vram gb. From 1 to 512. Default 80.0. |
| run_id | STRING | run | Run id. Default run. |
| hard_stopopt | BOOLEAN | false | Hard stop. Default False. |
| probe_gbopt | FLOAT | -1.0-1–1024 | Probe gb. From -1 to 1024. Default -1.0. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |
| latent | LATENT | — |
| vram_usage_gb | FLOAT | — |