mirror of
https://github.com/Comfy-Org/desktop.git
synced 2026-11-04 23:47:43 +00:00
[GH-ISSUE #1745] [Bug]: Windows Desktop crashes on AMD RX 6900 XT during torch.cuda.is_available() with PyTorch ROCm 2.9.1+rocm7.2.1 #542
Labels
No labels
area: desktop-app
Bounty
bug
bug-cop:answered
bug-cop:response-received
component: installer
component: preload
custom node
dependencies
Design
documentation
duplicate
enhancement
expected
frontend
good first issue
installation
Low-Priority
performance
Planned
platform: cross-platform
platform: windows
Potential Bug
Potential Bug
pull-request
question
refactor
Release
stale
User Support
V1
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
Comfy-Org/desktop#542
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 @massimomazzariol on GitHub (Jun 21, 2026).
Original GitHub issue: https://github.com/Comfy-Org/desktop/issues/1745
App Version
ComfyUI Desktop version: not sure / not available from the current logs.
Environment details from startup log:
Expected Behavior
ComfyUI Desktop should either:
Direct PyTorch verification should also return a normal result. In particular:
python -c "import torch; print(torch.cuda.is_available())"
should return either True or False, not crash or exit without printing a value
Actual Behavior
ComfyUI Desktop crashes during startup before the UI becomes usable.
The crash happens during GPU/backend detection, around:
torch.cuda.is_available()
The startup log reports:
Windows fatal exception: access violation
Direct verification inside the ComfyUI Desktop virtual environment also behaves incorrectly:
2.9.1+rocm7.2.1
7.2.53211-158bd99533
None
However:
python -c "import torch; print(torch.cuda.is_available())"
does not print True or False. It only prints:
[WARNING] failed to run offload-arch: binary not found.
and then returns to the prompt.
Also:
python -m torch.utils.collect_env
starts collecting environment information but does not complete with the expected environment report
Steps to Reproduce
Debug Logs
Browser Logs
Not available since the crash happens during backend startup before the ComfyUI web interface becomes usable, so I cannot collect browser console logs from the UI...
Settings JSON
No response
Other
This may be related to AMD/RDNA2 + Windows + bundled ROCm PyTorch inside ComfyUI Desktop.
The virtual environment contains a ROCm/HIP PyTorch build:
However, GPU detection does not complete correctly. The command:
python -c "import torch; print(torch.cuda.is_available())"
does not return True or False, and ComfyUI Desktop crashes during startup with a Windows fatal access violation in torch.cuda.is_available().
It would be useful if ComfyUI Desktop could either support this AMD ROCm/RDNA2 setup correctly or detect the unsupported/broken configuration and show a clear error message instead of crashing.