mirror of
https://github.com/Comfy-Org/comfy-aimdo.git
synced 2026-11-04 23:37:21 +00:00
[GH-ISSUE #5] DynamicVRAM (comfy-aimdo 0.2.1) fails to initialize on Windows with PyTorch 2.10.0+cu130 (CUDA 13), falling back to legacy ModelPatcher #3
Labels
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
Comfy-Org/comfy-aimdo#3
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @phazei on GitHub (Feb 25, 2026).
Original GitHub issue: https://github.com/Comfy-Org/comfy-aimdo/issues/5
Environment
Launch Args
Issue
ComfyUI logs:
comfy_aimdoimports successfully in Python, and CUDA12.612.8 (I just updated that, I have 12.8 and 13.0) runtime DLL is resolvable via PATH. DynamicVRAM still does not initialize.I'm not sure what else to try
@phazei commented on GitHub (Feb 25, 2026):
(venv) D:\AITools\StabilityMatrixData\Packages\ComfyUI>pip show comfy-aimdo
Name: comfy-aimdo
Version: 0.2.1
Summary: AI Model Dynamic Offloader for ComfyUI
Home-page:
Author: rattus
Author-email:
License:
Location: D:\AITools\StabilityMatrixData\Packages\ComfyUI\venv\Lib\site-packages
Requires:
Required-by:
(venv) D:\AITools\StabilityMatrixData\Packages\ComfyUI>python -c "import comfy_aimdo, comfy_aimdo.control as c; print('import OK'); print('cuda:', import('torch').version.cuda); print('control file:', c.file)"
import OK
cuda: 13.0
control file: D:\AITools\StabilityMatrixData\Packages\ComfyUI\venv\Lib\site-packages\comfy_aimdo\control.py
(venv) D:\AITools\StabilityMatrixData\Packages\ComfyUI>python -c "from comfy_aimdo import control; print(control.init())"
True
(venv) D:\AITools\StabilityMatrixData\Packages\ComfyUI>python -c "import comfy.model_management as m; print('dynamic_vram:', getattr(m, 'dynamic_vram', None))"
dynamic_vram: None
I tried launching with
--fast dynamic_vram --disable-cuda-mallocas well with no change@phazei commented on GitHub (Feb 25, 2026):
(venv) D:\AITools\StabilityMatrixData\Packages\ComfyUI>python -c "import comfy_aimdo.control as c, torch; print(c.init()); print(c.init_device(torch.cuda.current_device()))"
True
aimdo: src-win/cuda-detour.c:35:ALL:aimdo_setup_hooks: cudart64_12.dll not foundFalse
(venv) D:\AITools\StabilityMatrixData\Packages\ComfyUI>where cudart64_12.dll
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8\bin\cudart64_12.dll
(venv) D:\AITools\StabilityMatrixData\Packages\ComfyUI>python -c "import ctypes; ctypes.CDLL(r'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8\bin\cudart64_12.dll'); print('loaded')"
loaded
@rattus128 commented on GitHub (Feb 25, 2026):
Thanks for the report. There is an issue in cuda13 + recent versions of Aimdo. Will post a fix shortly.
@rattus128 commented on GitHub (Feb 25, 2026):
When comfy Aimdo 0.2.2 filters through this should be fixed.