If the Nvidia driver was updated, the tool should not attempt to do a fresh reinstall of the run folder.

It is only if the base container has changed that a new venv should be created

PyTorch upgrade if a new version is available and you have DISABLE_UPGRADES disabled.

When it installs, the == Installing torch using uv with backend: cu130 shows the backend used.

I expect what is happening is that a new version of Torch was released, but the corresponding TorchAudio was not yet and the restart caused the issues you are facing.

The solution would likely be to use the TORCH_LOCK environment variable (which is not enabled by default) and specify the exact version of all 3x primary packages to install. For example:

TORCH_LOCK=torch==2.9.1+cu130 torchvision==0.24.1+cu130 torchaudio==2.9.1+cu130

I run the CUDA 13.2 version, so if I do:

# Adapt your container name
% docker exec -it comfyui-nvidia /bin/bash
% sudo su -l comfy
# As comfy
% source /comfy/mnt/venv/bin/activate
% pip freeze | grep torch

I get

torch==2.12.0+cu130
torchaudio==2.11.0+cu130
[...]
torchvision==0.27.0+cu130

giving me the values which I could use with TORCH_LOCKenvironment variable.

The fact that a TorchAudio release is out of sync with Torch is unfortunately outside of my control.

I would usually check the table on:

GitHub

GitHub - pytorch/vision: Datasets, Transforms and Models...

Datasets, Transforms and Models specific to Computer Vision - pytorch/vision

https://github.com/pytorch/audio

From this research I see that Audio is going in maintenance mode but it is still a dependency for ComfyUI per https://github.com/Comfy-Org/ComfyUI/blob/master/requirements.txt

I see: TorchAudio 2.11 "this release is compatible with torch 2.11 and is compatible with future versions of torch."

I think I see a potential issue:"Introducing CUDA 13.2 and Deprecating CUDA 12.8 (Release 2.12)"

https://dev-discuss.pytorch.org/t/introducing-cuda-13-2-and-deprecating-cuda-12-8-release-2-12/3337

Good news is that cu132 finally has it own download page

https://download.pytorch.org/whl/cu132

Bad news ... uv does not yet support it, so not moving yet

9 hours ago, martial said:

If the Nvidia driver was updated, the tool should not attempt to do a fresh reinstall of the run folder.

It is only if the base container has changed that a new venv should be created

PyTorch upgrade if a new version is available and you have DISABLE_UPGRADES disabled.

When it installs, the == Installing torch using uv with backend: cu130 shows the backend used.

I expect what is happening is that a new version of Torch was released, but the corresponding TorchAudio was not yet and the restart caused the issues you are facing.

The solution would likely be to use the TORCH_LOCK environment variable (which is not enabled by default) and specify the exact version of all 3x primary packages to install. For example:

TORCH_LOCK=torch==2.9.1+cu130 torchvision==0.24.1+cu130 torchaudio==2.9.1+cu130

I run the CUDA 13.2 version, so if I do:

# Adapt your container name
% docker exec -it comfyui-nvidia /bin/bash
% sudo su -l comfy
# As comfy
% source /comfy/mnt/venv/bin/activate
% pip freeze | grep torch

I get

torch==2.12.0+cu130
torchaudio==2.11.0+cu130
[...]
torchvision==0.27.0+cu130

giving me the values which I could use with TORCH_LOCKenvironment variable.

The fact that a TorchAudio release is out of sync with Torch is unfortunately outside of my control.

I would usually check the table on:

GitHub

GitHub - pytorch/vision: Datasets, Transforms and Models...

Datasets, Transforms and Models specific to Computer Vision - pytorch/vision

https://github.com/pytorch/audio

From this research I see that Audio is going in maintenance mode but it is still a dependency for ComfyUI per https://github.com/Comfy-Org/ComfyUI/blob/master/requirements.txt

Dockers are not my strong suit. If i am understanding correctly in unraid the environmental variable "Key" would be TORCH_LOCK when adding the variable? If that is the case would the entire "TORCH_LOCK=torch==2.9.1+cu130 torchvision==0.24.1+cu130 torchaudio==2.9.1+cu130" be the variable itself or would i be dropping the TORCH_LOCK as that is already set as the "key" in the templates page? Sorry if this seems like a really dumb question, i just want to try my best at getting this to work.

Edited June 3Jun 3 by sentein
Double Quote - Fixed

I run the ubuntu24_cuda12.9 version on Unraid.

It uses:

torch==2.11.0+cu129

torchaudio==2.11.0+cu129

torchvision==0.26.0+cu129

From the "Docker" tab, click on the application icon and use "Edit"

So I could add "Add another Path, Port, Variable, Label or Device" with the following parameters:

Edited June 3Jun 3 by martial

3 minutes ago, martial said:

I run the ubuntu24_cuda12.9 version on Unraid.

It uses:

torch==2.11.0+cu129

torchaudio==2.11.0+cu129

torchvision==0.26.0+cu129

From the "Docker" tab, click on the application icon and use "Edit"

So I could add "Add another Path, Port, Variable, Label or Device" with the following parameters:

Thats how i thought it might layout. I will give this a try tonight. Thank you for the assistance. I don't want to pay for another unraid license unless i can get this to work. I already have too many.

Keep in mind if you use the template that you can select the CUDA version you want to run against.

I have made 12.8 the default but with the depreciation of 12.8 for PyTorch I might be forced to make 12.9 the default "latest" shortly.

It is recommended to manually select which version of CUDA you want to run against (if your Unraid Nvidia driver is up to date you should be able to run 13.2 ). You can find it from the same "edit" page as "Repository": The list of possible values is in the table at https://github.com/mmartial/ComfyUI-Nvidia-Docker

Edited June 3Jun 3 by martial

1 hour ago, martial said:

Keep in mind if you use the template that you can select the CUDA version you want to run against.

I have made 12.8 the default but with the depreciation of 12.8 for PyTorch I might be forced to make 12.9 the default "latest" shortly.

It is recommended to manually select which version CUDA you want to run against (if your Unraid Nvidia driver is up to date you should be able to run 13.2 ). You can find it from the same "edit" page as "Repository": The list of possible values is in the table at https://github.com/mmartial/ComfyUI-Nvidia-Docker

Uploading Attachment...

I did this and loaded 12.8. it installed 13.0 instead. No idea why or if there is something in the docker base infostructure i need to erase to get the version matching to work properly. Any info on that would be helpful as well.

The image uploaded should have the right info (apparently it did not copy):

Beyond that you should have nothing to do, the tool create a venv specific to the "ubuntu24-cuda12.9" at first start.

7 hours ago, martial said:

The image uploaded should have the right info (apparently it did not copy):

Beyond that you should have nothing to do, the tool create a venv specific to the "ubuntu24-cuda12.9" at first start.

I used all of the above info. The below is the last lines in the logterminal for ComfyUI-Nvidia-Docker.

fo, av.bitstream, av.device, av.video.codeccontext, av.subtitles.stream, cffibackend (total: 198)

/comfyui-nvidia_init.bash: line 935: 445 Illegal instruction ${COMFY_CMDLINE_BASE} ${COMFY_CMDLINE_EXTRA}

!! ERROR: ComfyUI failed or exited with an error

!! Exiting script (ID: 100)

!! ERROR: subscript failed

!! Exiting script (ID: 1)

Below is the last lines in the log file in the user folder. Not really sure what the heck i am doing wrong. I guess if you have any info as to how i could screw this up so bad, please let me know.

[PRE] ComfyUI-Manager

[2026-06-03 18:22:01.217] WARNING: You need pytorch with cu130 or higher to use optimized CUDA operations.

[2026-06-03 18:22:01.218] Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'apply_rope_split_half', 'apply_rope_split_half1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}

[2026-06-03 18:22:01.218] Found comfy_kitchen backend cuda: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'apply_rope_split_half', 'apply_rope_split_half1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'gemv_awq_w4a16', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8']}

[2026-06-03 18:22:01.218] Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'apply_rope_split_half', 'apply_rope_split_half1', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'gemv_awq_w4a16', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8']}

[2026-06-03 18:22:01.271] Checkpoint files will always be loaded safely.

[2026-06-03 18:22:01.620] Total VRAM 24126 MB, total RAM 257907 MB

[2026-06-03 18:22:01.621] pytorch version: 2.11.0+cu129

[2026-06-03 18:22:01.622] Set vram state to: NORMAL_VRAM

[2026-06-03 18:22:01.623] Device: cuda:0 NVIDIA GeForce RTX 3090 : cudaMallocAsync

[2026-06-03 18:22:01.624] Using async weight offloading with 2 streams

[2026-06-03 18:22:01.624] Enabled pinned memory 232116.0

[2026-06-03 18:22:03.800] Using pytorch attention

[2026-06-03 18:22:35.846] DynamicVRAM support detected and enabled

[2026-06-03 18:22:35.846] Python version: 3.12.3 (main, Mar 23 2026, 19:04:32) [GCC 13.3.0]

[2026-06-03 18:22:35.847] ComfyUI version: 0.23.0

[2026-06-03 18:22:35.849] comfy-aimdo version: 0.4.8

[2026-06-03 18:22:35.852] comfy-kitchen version: 0.2.10

[2026-06-03 18:22:35.866] comfyui-frontend-package version: 1.44.19

[2026-06-03 18:22:35.867] comfyui-workflow-templates version: 0.9.92

[2026-06-03 18:22:35.867] comfyui-embedded-docs version: 0.5.2

[2026-06-03 18:22:35.867] comfy-kitchen version: 0.2.10

[2026-06-03 18:22:35.868] comfy-aimdo version: 0.4.8

[2026-06-03 18:22:35.874] [Prompt Server] web root: /comfy/mnt/venv/lib/python3.12/site-packages/comfyui_frontend_package/static

[2026-06-03 18:22:35.877] Asset seeder disabled

[2026-06-03 18:22:35.877] [START] ComfyUI-Manager

[2026-06-03 18:22:36.247] [ComfyUI-Manager] Using GitPython backend

[2026-06-03 18:22:36.413] [ComfyUI-Manager] network_mode: personal_cloud

That error just tells us that ComfyUI exited with an error.

Everything in the second log is just normal ComfyUI starting with PyTorch 2.11.0+cu129 and Comfy 0.23.0

"when" it fails is shows the line that caused the error, here the call to Comfy which is normal.

If you are comfortable with this, can you delete the container (do not delete the folder where your files are in /mnt/usr/appdata).

Then move the run folder that should be in /mnt/user/appdata/comfyui-nvidia/mnt to "mnt.old" (not the basedir, it contains your generations and models)

From that Community Apps, reinstall a new ComfyUI, use at least CUDA 12.9 and it should recreate the "mnt" folder with a fresh installation?

Importantly it also tells us that Comfy started so your installation is (from the view of the container's purpose) clean

Yes i will run that tonight. i have been deleting the container and the the appdata folder completely on every installation. The errors you saw above were from installing the CUDA 12.9 release with the TORCH_LOCK variable enabled as you described in your previous posts. At this point in the servers life i am "comfortable" with whatever. There is no permanent data i am going to lose. This was an attempt to make a "AI" server. LLMs Stable diffusion and ComfyUI as the main use cases. If i can get all of those to work i will be adding Shinobi to the mix. But as of right now i could erase the entire thing and not hesitate about it. Thank you for the help, I really do not understand why this is fighting me so much.

The thing is that it should not be fighting you at all ... I run 3x instances of the container with my primary on an RTX 4090 Linux Mint using the CUDA 13.2 base and it works fine.

The container's part appear to be working: it installs ComfyUI, then starts it, it is ComfyUI that then crash (I did not see anything wrong in the logs that you shared with us).

The 3090 supports the latest driver with CUDA 13.2 I believe. If you ssh into your Unraid --or use the >_ terminal-- and type nvidia-smi you will see by the top the max version of CUDA that the host (and therefore container runtime) can support, my Unraid runs Driver Version: 595.71.05 CUDA Version: 13.2 (ie I have not yet upgraded to the 610.43.02 driver that appears to be available) so I can use the 13.2 version (and do not use TORCH_LOCK)

From the release notes, I do not see any expected issue but maybe others that have a more recent driver can say what they see?

https://forums.developer.nvidia.com/t/610-release-feedback-discussion/371356

In the Settings/nvidia-driver page of my Unraid, I see in the GPU Driver Support a note:

Supported candidates: 580.159.03, 590.48.01, 595.71.05, 595.80, 610.43.02
Recommended Driver: 595.71.05 (open-source)

Apparently I am running the expected "recommended" driver.

nothing is working. Thanks for the help. I am going to have to move on i think.

2 minutes ago, sentein said:

nothing is working. Thanks for the help. I am going to have to move on i think.

Understood, maybe other will encounter the same issue and report their solution. I do not have the issue or am able to reproduce it currently, so I can not help further.

Is everyone having a problem on 7.3.1?

I'm having a strange issue as well, using Cuda 12.8 and a fresh install to see if it fixes comfyui but it did not. The error is the same as above I believe:
!! ERROR: ComfyUI failed or exited with an error

!! Exiting script (ID: 60)

!! ERROR: subscript failed

!! Exiting script (ID: 1)

Never had any issues a bit ago and my unraid version is on 7.0.0

And you are not using TORCH_LOCK, correct?

In your case, my belief is that Torch is trying to update to 2.12 but as noted above, they dropped support for CUDA 12.8 in Torch 2.12

Details: "Introducing CUDA 13.2 and Deprecating CUDA 12.8 (Release 2.12)" https://dev-discuss.pytorch.org/t/introducing-cuda-13-2-and-deprecating-cuda-12-8-release-2-12/3337

The two options I would recommend are either:

1) use TORCH_LOCK=torch==2.9.1+cu130 torchvision==0.24.1+cu130 torchaudio==2.9.1+cu130or a more recent Torch (as long as you get the right combination, I am not sure what it is honestly :) )

2) change the base container from ubuntu24-cuda12.8 to a more recent one.

Important note: latest is likely going to move from CUDA 12.8 to CUDA 12.9 because of that (likely this week end)

Edited June 5Jun 5 by martial

On 6/2/2026 at 10:05 AM, turtleinarock said:

Also having issues with

Unraid 7.3.1
Nvidia Driver Plugin: 2026.05.07
Nvidia Driver Version: 595.80
Image Tag: comfyui-nvidia-docker:ubuntu24_cuda13.0-latest and ..._cuda13.2-latest

I get an error during container start:

== Pre-installing/Upgrading torch
== Installing torch using uv with backend: cu130
Installing: torch torchvision torchaudio
!! ERROR: Torch installation failed
!! Exiting script (ID: 99)
!! ERROR: subscript failed
!! Exiting script (ID: 1)

I just tried starting up the container again and it worked just fine. Cuda 13.2 tag. No torch lock.

Edited June 5Jun 5 by turtleinarock

I am currently getting this error:

Traceback (most recent call last):
  File "/comfy/mnt/ComfyUI/./main.py", line 219, in <module>
    import execution
  File "/comfy/mnt/ComfyUI/execution.py", line 22, in <module>
    from latent_preview import set_preview_method
  File "/comfy/mnt/ComfyUI/latent_preview.py", line 5, in <module>
    from comfy.sd import VAE
  File "/comfy/mnt/ComfyUI/comfy/sd.py", line 14, in <module>
    import comfy.ldm.lightricks.vae.audio_vae
  File "/comfy/mnt/ComfyUI/comfy/ldm/lightricks/vae/audio_vae.py", line 5, in <module>
    import torchaudio
  File "/comfy/mnt/venv/lib/python3.12/site-packages/torchaudio/__init__.py", line 7, in <module>
    from . import _extension  # noqa  # usort: skip
    ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfy/mnt/venv/lib/python3.12/site-packages/torchaudio/_extension/__init__.py", line 36, in <module>
    _check_cuda_version()
  File "/comfy/mnt/venv/lib/python3.12/site-packages/torchaudio/_extension/utils.py", line 124, in _check_cuda_version
    raise RuntimeError(
RuntimeError: Detected that PyTorch and TorchAudio were compiled with different CUDA versions. PyTorch has CUDA version 13.0 whereas TorchAudio has CUDA version 12.9. Please install the TorchAudio version that matches your PyTorch version.
  ++ Attempting to recursively set ownership of /basedir/models/hypernetworks to 99:100 (might take a long time)
    -- /basedir/models/latent_upscale_models already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/latent_upscale_models to 99:100 (might take a long time)
    -- /basedir/models/loras already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/loras to 99:100 (might take a long time)
    -- /basedir/models/model_patches already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/model_patches to 99:100 (might take a long time)
    -- /basedir/models/optical_flow already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/optical_flow to 99:100 (might take a long time)
    -- /basedir/models/photomaker already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/photomaker to 99:100 (might take a long time)
    -- /basedir/models/style_models already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/style_models to 99:100 (might take a long time)
    -- /basedir/models/text_encoders already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/text_encoders to 99:100 (might take a long time)
    -- /basedir/models/unet already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/unet to 99:100 (might take a long time)
    -- /basedir/models/upscale_models already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/upscale_models to 99:100 (might take a long time)
    -- /basedir/models/vae already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/vae to 99:100 (might take a long time)
    -- /basedir/models/vae_approx already exists, skipping
  ++ Attempting to recursively set ownership of /basedir/models/vae_approx to 99:100 (might take a long time)
  ++ Creating /comfy/mnt/ComfyUI/user
!! COMFY_CMDLINE_EXTRA extended, make sure to use it in user script (if any):  --base-directory /basedir
== Using new ComfyUI Manager's required command line addition: --enable-manager
!! COMFY_CMDLINE_EXTRA extended, make sure to use it in user script (if any):  --base-directory /basedir --enable-manager

How can I resolve it? running a RTX PRO 4000 Blackwell and driver 595

The error I see in there is RuntimeError: Detected that PyTorch and TorchAudio were compiled with different CUDA versions. PyTorch has CUDA version 13.0 whereas TorchAudio has CUDA version 12.9. Please install the TorchAudio version that matches your PyTorch version.

You said you have a Blackwell card. I would try with the CUDA 13.2 container, not the 13.0 (or are you trying 12.9)?