[GH-ISSUE #1698] Unloading models on AMD #737

New issue

Closed

opened 2026-10-27 21:21:36 +00:00 by bsakaguchi · 1 comment

bsakaguchi commented

2026-10-27 21:21:36 +00:00

Owner

Originally created by @Lonmine on GitHub (Mar 28, 2025).
Original GitHub issue: https://github.com/Comfy-Org/ComfyUI-Manager/issues/1698

The Unload Models feature here don't seem to actually unload the checkpoints from the GPU memory, i imagine it is because it is an AMD gpu and not a Nvidea gpu, or it just leaves it on RAM, i don't have much RAM so when SDXL models load when another one is already on the memory comfyui crashes ( Exception Code: 0xC0000005), is there a workaround it?

Originally created by @Lonmine on GitHub (Mar 28, 2025). Original GitHub issue: https://github.com/Comfy-Org/ComfyUI-Manager/issues/1698 The Unload Models feature here don't seem to actually unload the checkpoints from the GPU memory, i imagine it is because it is an AMD gpu and not a Nvidea gpu, or it just leaves it on RAM, i don't have much RAM so when SDXL models load when another one is already on the memory comfyui crashes ( Exception Code: 0xC0000005), is there a workaround it?

bsakaguchi closed this issue

2026-10-27 21:21:36 +00:00

bsakaguchi commented

2026-10-27 21:21:36 +00:00

Author

Owner

@ltdrdata commented on GitHub (Mar 28, 2025):

Actually, that feature belongs to the ComfyUI core.
ComfyUI-Manager merely calls the unloading API.

This issue should be addressed in the core repo.

@ltdrdata commented on GitHub (Mar 28, 2025): Actually, that feature belongs to the ComfyUI core. ComfyUI-Manager merely calls the unloading API. This issue should be addressed in the core repo.

bsakaguchi referenced this issue

2026-10-27 21:32:02 +00:00

[PR #737] [MERGED] ComfyUI-NodeJS Node #1493