comfyanonymous
ComfyUI crashes instantly after clicking “Generate” on Windows 11 with ROCm + RX 6600 XT
Custom Node Testing
- [ ] I have tried disabling custom nodes and the issue persists (see how to disable custom nodes if you need help)
Your question
I successfully installed the new ROCm-enabled PyTorch on Windows 11 and everything loads fine in ComfyUI — the GPU is detected, models load, and there are no errors during startup.
However, the moment I click Generate, ComfyUI immediately crashes with no error message (the terminal just closes). This only happens during inference; model loading works normally.
Here’s my setup and logs:
GPU: AMD Radeon RX 6600 XT (gfx1032)
Driver: latest AMD ROCm PyTorch driver
PyTorch: 2.9.0+rocmsdk20251116
ROCm runtime: 7.1.52802
OS: Windows 11 Pro
Python: 3.12 (Anaconda)
ComfyUI: v0.3.76
torch.cuda.is_available() → True
Device name is correctly detected
I also tested:
HSA_OVERRIDE_GFX_VERSION=10.3.0 → same crash
Has anyone with a 6600 XT on Windows + ROCm gotten ComfyUI inference to run without crashing? Is this a ROCm issue or a ComfyUI compatibility problem with RDNA2?
Any help or debugging tips would be appreciated.
Full logs :
(base) C:\Users\merfk>conda activate comfyui_rocm
(comfyui_rocm) C:\Users\merfk>python -c "import torch" 2>nul && echo Success || echo Failure
Success
(comfyui_rocm) C:\Users\merfk>python -c "import torch; print(torch.cuda.is_available())"
True
(comfyui_rocm) C:\Users\merfk>python -c "import torch; print(f'device name [0]:', torch.cuda.get_device_name(0))"
device name [0]: AMD Radeon RX 6600 XT
(comfyui_rocm) C:\Users\merfk>python -m torch.utils.collect_env
Collecting environment information...
PyTorch version: 2.9.0+rocmsdk20251116
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: 7.1.52802-561cc400e1
OS: Microsoft Windows 11 Pro (10.0.26200 64-bit)
GCC version: Could not collect
Clang version: Could not collect
CMake version: Could not collect
Libc version: N/A
Python version: 3.12.12 | packaged by Anaconda, Inc. | (main, Oct 21 2025, 20:05:38) [MSC v.1929 64 bit (AMD64)] (64-bit runtime)
Python platform: Windows-11-10.0.26200-SP0
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to:
GPU models and configuration: AMD Radeon RX 6600 XT (gfx1032)
Nvidia driver version: Could not collect
cuDNN version: Could not collect
Is XPU available: False
HIP runtime version: 7.1.52802
MIOpen runtime version: 3.5.1
Is XNNPACK available: True
CPU:
Name: Intel(R) Core(TM) i5-10400F CPU @ 2.90GHz
Manufacturer: GenuineIntel
Family: 205
Architecture: 9
ProcessorType: 3
DeviceID: CPU0
CurrentClockSpeed: 2904
MaxClockSpeed: 2904
L2CacheSize: 1536
L2CacheSpeed: None
Revision: None
Versions of relevant libraries:
[pip3] numpy==2.3.5
[pip3] torch==2.9.0+rocmsdk20251116
[pip3] torchaudio==2.9.0+rocmsdk20251116
[pip3] torchsde==0.2.6
[pip3] torchvision==0.24.0+rocmsdk20251116
[conda] numpy 2.3.5 pypi_0 pypi
[conda] torch 2.9.0+rocmsdk20251116 pypi_0 pypi
[conda] torchaudio 2.9.0+rocmsdk20251116 pypi_0 pypi
[conda] torchsde 0.2.6 pypi_0 pypi
[conda] torchvision 0.24.0+rocmsdk20251116 pypi_0 pypi
(comfyui_rocm) C:\Users\merfk>cd Comfyui
(comfyui_rocm) C:\Users\merfk\ComfyUI>
(comfyui_rocm) C:\Users\merfk\ComfyUI>set HSA_OVERRIDE_GFX_VERSION=10.3.0
(comfyui_rocm) C:\Users\merfk\ComfyUI>python main.py
Checkpoint files will always be loaded safely.
Total VRAM 8176 MB, total RAM 32667 MB
pytorch version: 2.9.0+rocmsdk20251116
Set: torch.backends.cudnn.enabled = False for better AMD performance.
AMD arch: gfx1032
ROCm version: (7, 1)
Set vram state to: NORMAL_VRAM
Device: cuda:0 AMD Radeon RX 6600 XT : native
Enabled pinned memory 14700.0
Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention
Python version: 3.12.12 | packaged by Anaconda, Inc. | (main, Oct 21 2025, 20:05:38) [MSC v.1929 64 bit (AMD64)]
ComfyUI version: 0.3.76
ComfyUI frontend version: 1.33.10
[Prompt Server] web root: C:\Users\merfk.conda\envs\comfyui_rocm\Lib\site-packages\comfyui_frontend_package\static
Total VRAM 8176 MB, total RAM 32667 MB
pytorch version: 2.9.0+rocmsdk20251116
Set: torch.backends.cudnn.enabled = False for better AMD performance.
AMD arch: gfx1032
ROCm version: (7, 1)
Set vram state to: NORMAL_VRAM
Device: cuda:0 AMD Radeon RX 6600 XT : native
Enabled pinned memory 14700.0
Import times for custom nodes:
0.0 seconds: C:\Users\merfk\ComfyUI\custom_nodes\websocket_image_save.py
Context impl SQLiteImpl.
Will assume non-transactional DDL.
No target revision found.
Starting server
To see the GUI go to: http://127.0.0.1:8188/
got prompt
model weight dtype torch.float16, manual cast: None
model_type EPS
Using split attention in VAE
Using split attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
Requested to load SDXLClipModel
loaded completely; 95367431640625005117571072.00 MB usable, 1560.80 MB loaded, full load: True
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cuda:0, dtype: torch.float16
Requested to load SDXLClipModel
(comfyui_rocm) C:\Users\merfk\ComfyUI>
Logs
Other
No response
The instant crash is usually related to the pagefile size. Try setting it "managed by the system" or increase the limit.
The instant crash is usually related to the pagefile size. Try setting it "managed by the system" or increase the limit.
Thanks for the suggestion, I tried increasing the pagefile size, but unfortunately the crash still happens immediately when I click Generate. (comfyui_rocm) C:\Users\merfk\ComfyUI>python main.py --use-split-cross-attention Checkpoint files will always be loaded safely. Total VRAM 8176 MB, total RAM 32667 MB pytorch version: 2.9.0+rocmsdk20251116 Set: torch.backends.cudnn.enabled = False for better AMD performance. AMD arch: gfx1032 ROCm version: (7, 1) Set vram state to: NORMAL_VRAM Device: cuda:0 AMD Radeon RX 6600 XT : native Enabled pinned memory 14700.0 Using split optimization for attention Python version: 3.12.12 | packaged by Anaconda, Inc. | (main, Oct 21 2025, 20:05:38) [MSC v.1929 64 bit (AMD64)] ComfyUI version: 0.3.76 ComfyUI frontend version: 1.33.10 [Prompt Server] web root: C:\Users\merfk.conda\envs\comfyui_rocm\Lib\site-packages\comfyui_frontend_package\static Total VRAM 8176 MB, total RAM 32667 MB pytorch version: 2.9.0+rocmsdk20251116 Set: torch.backends.cudnn.enabled = False for better AMD performance. AMD arch: gfx1032 ROCm version: (7, 1) Set vram state to: NORMAL_VRAM Device: cuda:0 AMD Radeon RX 6600 XT : native Enabled pinned memory 14700.0
Import times for custom nodes: 0.0 seconds: C:\Users\merfk\ComfyUI\custom_nodes\websocket_image_save.py
Context impl SQLiteImpl. Will assume non-transactional DDL. No target revision found. Starting server
To see the GUI go to: http://127.0.0.1:8188 got prompt model weight dtype torch.float16, manual cast: None model_type EPS Using split attention in VAE Using split attention in VAE VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16 Requested to load SDXLClipModel loaded completely; 95367431640625005117571072.00 MB usable, 1560.80 MB loaded, full load: True CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cuda:0, dtype: torch.float16 Requested to load SDXLClipModel
(comfyui_rocm) C:\Users\merfk\ComfyUI>
where did you get this rocm that has 6600xt or any rdna2 is recognised and working with in windows ?
where did you get this rocm that has 6600xt or any rdna2 is recognised and working with in windows ? I used those guides https://www.amd.com/en/resources/support-articles/release-notes/RN-AMDGPU-WINDOWS-PYTORCH-7-1-1.html
https://rocm.docs.amd.com/projects/radeon-ryzen/en/latest/docs/install/installrad/windows/install-pytorch.html
And I saw a post in your reddit that said this:
just regular windows 11. pytorch instructions are here: https://github.com/ROCm/TheRock/tree/main/external-builds/pytorch#build-instructions . you should be able to use to use rocm wheels from here (need all 4): https://rocm.nightlies.amd.com/v2/gfx103X-dgpu/ pytorch took me ~2h to compile and you may need to install some of the dependencies from here: https://github.com/ROCm/TheRock/blob/main/docs/development/windows_support.md#building-therock-from-source . if rocm wheels are not compatible with your gpu, try using env. variable HSA_OVERRIDE_GFX_VERSION="10.3.0"
(I didn’t follow that guide Should I have followed that guide?)
I have tried compiling for a full day and always some problems ,as far as I know unless you do extensive patching and manual fixing they still do not work for 103x, there is some things happening though in their github
I have tried compiling for a full day and always some problems ,as far as I know unless you do extensive patching and manual fixing they still do not work for 103x, there is some things happening though in their github
How long do you think it will take for ROCm on Windows to become fully stable? And do you think it will properly support the RX 6000 series, or should we just wait for newer ZLUDA releases instead?
I have tried compiling for a full day and always some problems ,as far as I know unless you do extensive patching and manual fixing they still do not work for 103x, there is some things happening though in their github
How long do you think it will take for ROCm on Windows to become fully stable? And do you think it will properly support the RX 6000 series, or should we just wait for newer ZLUDA releases instead?
they are getting there, follow this directory, when there is torch / torchvision / torchaudio added and it includes windows packages it would mean it is done .
https://rocm.nightlies.amd.com/v2/gfx103X-dgpu/