[GH-ISSUE #1300] Pytorch support for Pascal GPUs

bsakaguchi commented

2026-10-31 22:12:03 +00:00

Owner

Originally created by @robhay2628 on GitHub (Sep 9, 2025).
Original GitHub issue: https://github.com/Comfy-Org/desktop/issues/1300

Prerequisites

I have searched the existing issues and checked the recent builds/commits
I have read the documentation and this feature is not already available
This is a single feature request (not multiple features combined)

Problem Statement

I was running Desktop ComfyUI fine until I had to reset my computer this past week. I went to run the new installation and I was getting a CUDA Core error. Apparently pytorch support for older GPUs like mine were removed in the updates. I can't download and older version other than a web based one. I was hoping you could put support back in for the GTX 1080 GPUs. It was slow, but it worked for my needs.

Proposed Solution

Seems like the installation software could be updated to select older GPUs for installations perhaps? Chat GTP gave a explanation of how to modify my installation, but it is beyond my programming understanding.

Potential workflow

Go to ....
Press ....
...

Impact Level

Critical (Blocking my work)

Alternatives Considered

No response

Additional information

No response

┆Issue is synchronized with this Notion page by Unito

Originally created by @robhay2628 on GitHub (Sep 9, 2025). Original GitHub issue: https://github.com/Comfy-Org/desktop/issues/1300 ### Prerequisites - [x] I have searched the existing issues and checked the recent builds/commits - [x] I have read the documentation and this feature is not already available - [x] This is a single feature request (not multiple features combined) ### Problem Statement I was running Desktop ComfyUI fine until I had to reset my computer this past week. I went to run the new installation and I was getting a CUDA Core error. Apparently pytorch support for older GPUs like mine were removed in the updates. I can't download and older version other than a web based one. I was hoping you could put support back in for the GTX 1080 GPUs. It was slow, but it worked for my needs. ### Proposed Solution Seems like the installation software could be updated to select older GPUs for installations perhaps? Chat GTP gave a explanation of how to modify my installation, but it is beyond my programming understanding. ### Potential workflow 1. Go to .... 2. Press .... 3. ... ### Impact Level Critical (Blocking my work) ### Alternatives Considered _No response_ ### Additional information _No response_ ┆Issue is synchronized with this [Notion page](https://www.notion.so/Issue-1300-Pytorch-support-for-Pascal-GPUs-2696d73d36508184a979f79e88e8dfb6) by [Unito](https://www.unito.io)

bsakaguchi added the

enhancement

label

2026-10-31 22:12:03 +00:00

bsakaguchi commented

2026-10-31 22:12:04 +00:00

Author

Owner

@webfiltered commented on GitHub (Sep 9, 2025):

Hey there, can definitely relate - I held onto my 1060 for years.

Why this happened

This issue is due to NVIDIA dropping support for the 10 series in CUDA 12.9. The latest pytorch uses it. We're looking at adding a way to detect the user's specific card model, so we can try to pick the right version of PyTorch to install.

In the meantime, you can fix this by changing the version installed into your python virtual environment. By design, once installed, the version is not automatically updated/downgraded (even when you update the desktop app).

If you can -start- the desktop app and get to the normal node graph interface, you can work around this from the in-app terminal.

Replace torch from inside the desktop app

Open the desktop app
Click the terminal icon >_ in the bottom left
Click TERMINAL
Run: uv pip uninstall torch torchvision torchaudio
Run: uv pip install torch torchvision torchaudio --index-url=https://download.pytorch.org/whl/cu128
IMPORTANT: Restart the app once the install has finished

This will (at time of writing) install pytorch 2.8 using CUDA 12.8, which still supports the 10 series.

Desktop app won't load

If you're -not- able to start the desktop app, you may need to use the full path to uv:

Open PowerShell and cd to Documents\ComfyUI (or the folder you selected during install)
Run steps 4 & 5 above, but replace uv at the start with: & "$env:LOCALAPPDATA\ComfyUI\assets\uv\win\uv.exe"

@webfiltered commented on GitHub (Sep 9, 2025): Hey there, can definitely relate - I held onto my 1060 for _years_. ### Why this happened This issue is due to NVIDIA dropping support for the 10 series in CUDA 12.9. The latest `pytorch` uses it. We're looking at adding a way to detect the user's specific card model, so we can try to pick the right version of PyTorch to install. In the meantime, you can fix this by changing the version installed into your python virtual environment. By design, once installed, the version is not automatically updated/downgraded (even when you update the desktop app). If you can -start- the desktop app and get to the normal node graph interface, you can work around this from the in-app terminal. ### Replace torch from inside the desktop app 1. Open the desktop app 2. Click the **terminal icon** `>_` in the bottom left 3. Click **TERMINAL** 4. Run: `uv pip uninstall torch torchvision torchaudio` 5. Run: `uv pip install torch torchvision torchaudio --index-url=https://download.pytorch.org/whl/cu128` 6. IMPORTANT: Restart the app once the install has finished This will (at time of writing) install pytorch 2.8 using CUDA 12.8, which still supports the 10 series. ### Desktop app won't load If you're -not- able to start the desktop app, you may need to use the full path to `uv`: 1. Open PowerShell and `cd` to `Documents\ComfyUI` (or the folder you selected during install) 2. Run steps 4 & 5 above, but replace `uv` at the start with: `& "$env:LOCALAPPDATA\ComfyUI\assets\uv\win\uv.exe"`

bsakaguchi commented

2026-10-31 22:12:04 +00:00

Author

Owner

@dosubot[bot] commented on GitHub (Dec 9, 2025):

Hi, @robhay2628. I'm Dosu, and I'm helping the desktop team manage their backlog and am marking this issue as stale.

Issue Summary:

You reported that recent updates using CUDA 12.9 dropped support for older Pascal GPUs like the GTX 1080, causing errors.
The root cause is NVIDIA dropping 10 series GPU support starting with CUDA 12.9.
A workaround was provided to manually downgrade PyTorch to version 2.8 with CUDA 12.8, restoring compatibility.
I suggested adding GPU detection to automatically install compatible PyTorch versions for different hardware.
The issue is considered resolved with the manual downgrade workaround in place.

Next Steps:

Please confirm if this issue is still relevant with the latest version of the desktop app.
If it is, feel free to keep the discussion open by commenting here; otherwise, this issue will be automatically closed in 7 days.

Thanks for your understanding and contribution!

@dosubot[bot] commented on GitHub (Dec 9, 2025): Hi, @robhay2628. I'm [Dosu](https://dosu.dev), and I'm helping the desktop team manage their backlog and am marking this issue as stale. **Issue Summary:** - You reported that recent updates using CUDA 12.9 dropped support for older Pascal GPUs like the GTX 1080, causing errors. - The root cause is NVIDIA dropping 10 series GPU support starting with CUDA 12.9. - A workaround was provided to manually downgrade PyTorch to version 2.8 with CUDA 12.8, restoring compatibility. - I suggested adding GPU detection to automatically install compatible PyTorch versions for different hardware. - The issue is considered resolved with the manual downgrade workaround in place. **Next Steps:** - Please confirm if this issue is still relevant with the latest version of the desktop app. - If it is, feel free to keep the discussion open by commenting here; otherwise, this issue will be automatically closed in 7 days. Thanks for your understanding and contribution!

bsakaguchi commented

2026-10-31 22:12:04 +00:00

Author

Owner

@weizhou0626 commented on GitHub (Dec 16, 2025):

I’m experiencing the same issue. I have a GTX 1060 6GB GPU. Although it’s quite old, it used to work fine. Since the update, it no longer works at all and I’ve encountered endless problems. Previously, it was just a bit slow but could still generate images. Now it fails completely.

@weizhou0626 commented on GitHub (Dec 16, 2025): I’m experiencing the same issue. I have a GTX 1060 6GB GPU. Although it’s quite old, it used to work fine. Since the update, it no longer works at all and I’ve encountered endless problems. Previously, it was just a bit slow but could still generate images. Now it fails completely.

bsakaguchi commented

2026-10-31 22:12:04 +00:00

Author

Owner

@dosubot[bot] commented on GitHub (Dec 16, 2025):

@robinjhuang, the user with a GTX 1060 6GB GPU is still experiencing the issue where the app fails completely after the update, whereas it used to work albeit slowly before. Could you please assist with this?

@dosubot[bot] commented on GitHub (Dec 16, 2025): @robinjhuang, the user with a GTX 1060 6GB GPU is still experiencing the issue where the app fails completely after the update, whereas it used to work albeit slowly before. Could you please assist with this?

bsakaguchi commented

2026-10-31 22:12:04 +00:00

Author

Owner

@robhay2628 commented on GitHub (Dec 16, 2025):

I managed to snag an older copy of ComfyUI that works with Pascal Cards, but GPU memory can be a problem. Just keep the pictures small. Here is a link to the file on my iCloud.

https://www.icloud.com/iclouddrive/03fZ6P6nNqief4fLwmv_9ElBQ#Comfy_UI_Related_Aps_and_Options

@robhay2628 commented on GitHub (Dec 16, 2025): I managed to snag an older copy of ComfyUI that works with Pascal Cards, but GPU memory can be a problem. Just keep the pictures small. Here is a link to the file on my iCloud. https://www.icloud.com/iclouddrive/03fZ6P6nNqief4fLwmv_9ElBQ#Comfy_UI_Related_Aps_and_Options

bsakaguchi commented

2026-10-31 22:12:04 +00:00

Author

Owner

@weizhou0626 commented on GitHub (Dec 17, 2025):

Why don't desktop developers allow users to select dependency versions based on their actual hardware? Forcing the use of the latest dependencies causes countless issues. I hope developers can quickly update desktop software to implement this feature. Additionally, the biggest obstacle to ComfyUI's development today is compatibility and the constant stream of problems, which is extremely frustrating for non-programming professionals. The ComfyUI team should establish a dedicated department to address the compatibility issues among its hundreds of dependencies.

@weizhou0626 commented on GitHub (Dec 17, 2025): Why don't desktop developers allow users to select dependency versions based on their actual hardware? Forcing the use of the latest dependencies causes countless issues. I hope developers can quickly update desktop software to implement this feature. Additionally, the biggest obstacle to ComfyUI's development today is compatibility and the constant stream of problems, which is extremely frustrating for non-programming professionals. The ComfyUI team should establish a dedicated department to address the compatibility issues among its hundreds of dependencies.

bsakaguchi commented

2026-10-31 22:12:05 +00:00

Author

Owner

@robhay2628 commented on GitHub (Dec 17, 2025):

It is not Comfy UIs problem. The tools given to them had an update to Pytorch, which handles the GPUs. Pytorch dropped Pascal based GPU support. I think it is NVIDA that writes the code for Pytorch which integrates into Python's Libraries. From what I hear it was NVIDIA that dropped support in Pytorch for the PASCAL based GPUs. The are simply making the 10 Serries cards obsolete with new software releases, thus moving development and the economic activity towards the newer cards.

@robhay2628 commented on GitHub (Dec 17, 2025): It is not Comfy UIs problem. The tools given to them had an update to Pytorch, which handles the GPUs. Pytorch dropped Pascal based GPU support. I think it is NVIDA that writes the code for Pytorch which integrates into Python's Libraries. From what I hear it was NVIDIA that dropped support in Pytorch for the PASCAL based GPUs. The are simply making the 10 Serries cards obsolete with new software releases, thus moving development and the economic activity towards the newer cards.

bsakaguchi commented

2026-10-31 22:12:05 +00:00

Author

Owner

@msnbcorp commented on GitHub (Feb 3, 2026):

Hello, I am agree with this bug report, I also have a GTX 1070 and it works fine with comfy UI with the work around, even with modern models like Z Turbo Image etc.
I think the detection and the automatic application of the workaround is not a very extensive modification ans is feasable.
Due to the price of ram that impact every component in the world, I think that keeping a maximum compatibilities with old hardware could be a good policy.

@msnbcorp commented on GitHub (Feb 3, 2026): Hello, I am agree with this bug report, I also have a GTX 1070 and it works fine with comfy UI with the work around, even with modern models like Z Turbo Image etc. I think the detection and the automatic application of the workaround is not a very extensive modification ans is feasable. Due to the price of ram that impact every component in the world, I think that keeping a maximum compatibilities with old hardware could be a good policy.

bsakaguchi commented

2026-10-31 22:12:05 +00:00

Author

Owner

@duffyjp commented on GitHub (Apr 24, 2026):

@msnbcorp So you did the steps from https://github.com/Comfy-Org/desktop/issues/1300#issuecomment-3269636287 and that worked? I tried it myself and I'm still getting that no kernel available stuff.

torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device
Search for `cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.

This morning I swapped out a 6gb 2060 for a 11gb 1080 Ti. I knew I was in for trouble with that, but it was $50 and games run WAY better.

(after the workaround & restart)

UserWarning: Found GPU0 NVIDIA GeForce GTX 1080 Ti which is of compute capability (CC) 6.1.
The following list shows the CCs this version of PyTorch was built for and the hardware CCs it supports:
- 7.5 which supports hardware CC >=7.5,<8.0
- 8.0 which supports hardware CC >=8.0,<9.0 except {8.7}
- 8.6 which supports hardware CC >=8.6,<9.0 except {8.7}
- 9.0 which supports hardware CC >=9.0,<10.0
- 10.0 which supports hardware CC >=10.0,<11.0 except {10.1}
- 12.0 which supports hardware CC >=12.0,<13.0
Please follow the instructions at https://pytorch.org/get-started/locally/ to install a PyTorch release that supports one of these CUDA versions: 12.6

Did you make any changes to @webfiltered's instructions?

@duffyjp commented on GitHub (Apr 24, 2026): @msnbcorp So you did the steps from https://github.com/Comfy-Org/desktop/issues/1300#issuecomment-3269636287 and that worked? I tried it myself and I'm still getting that no kernel available stuff. ``` torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device Search for `cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information. ``` This morning I swapped out a 6gb 2060 for a 11gb 1080 Ti. I knew I was in for trouble with that, but it was $50 and games run WAY better. (after the workaround & restart) ``` UserWarning: Found GPU0 NVIDIA GeForce GTX 1080 Ti which is of compute capability (CC) 6.1. The following list shows the CCs this version of PyTorch was built for and the hardware CCs it supports: - 7.5 which supports hardware CC >=7.5,<8.0 - 8.0 which supports hardware CC >=8.0,<9.0 except {8.7} - 8.6 which supports hardware CC >=8.6,<9.0 except {8.7} - 9.0 which supports hardware CC >=9.0,<10.0 - 10.0 which supports hardware CC >=10.0,<11.0 except {10.1} - 12.0 which supports hardware CC >=12.0,<13.0 Please follow the instructions at https://pytorch.org/get-started/locally/ to install a PyTorch release that supports one of these CUDA versions: 12.6 ``` Did you make any changes to @webfiltered's instructions?

bsakaguchi commented

2026-10-31 22:12:05 +00:00

Author

Owner

@robhay2628 commented on GitHub (Apr 24, 2026):

@msnbcorp So you did the steps from #1300 (comment) and that worked? I tried it myself and I'm still getting that no kernel available stuff.

torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device
Search for `cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.

This morning I swapped out a 6gb 2060 for a 11gb 1080 Ti. I knew I was in for trouble with that, but it was $50 and games run WAY better.

(after the workaround & restart)

UserWarning: Found GPU0 NVIDIA GeForce GTX 1080 Ti which is of compute capability (CC) 6.1.
The following list shows the CCs this version of PyTorch was built for and the hardware CCs it supports:
- 7.5 which supports hardware CC >=7.5,<8.0
- 8.0 which supports hardware CC >=8.0,<9.0 except {8.7}
- 8.6 which supports hardware CC >=8.6,<9.0 except {8.7}
- 9.0 which supports hardware CC >=9.0,<10.0
- 10.0 which supports hardware CC >=10.0,<11.0 except {10.1}
- 12.0 which supports hardware CC >=12.0,<13.0
Please follow the instructions at https://pytorch.org/get-started/locally/ to install a PyTorch release that supports one of these CUDA versions: 12.6

Did you make any changes to @webfiltered's instructions?

Again, I managed to find an older copy of Comfy UI, that works for Pascal based GPUs if that is what you are trying to doi here, such as the 1080 cards. Here is the link.
https://www.icloud.com/iclouddrive/03fZ6P6nNqief4fLwmv_9ElBQ#Comfy_UI_Related_Aps_and_Options

@robhay2628 commented on GitHub (Apr 24, 2026): > [@msnbcorp](https://github.com/msnbcorp) So you did the steps from [#1300 (comment)](https://github.com/Comfy-Org/desktop/issues/1300#issuecomment-3269636287) and that worked? I tried it myself and I'm still getting that no kernel available stuff. > > ``` > torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device > Search for `cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information. > ``` > > This morning I swapped out a 6gb 2060 for a 11gb 1080 Ti. I knew I was in for trouble with that, but it was $50 and games run WAY better. > > (after the workaround & restart) > > ``` > UserWarning: Found GPU0 NVIDIA GeForce GTX 1080 Ti which is of compute capability (CC) 6.1. > The following list shows the CCs this version of PyTorch was built for and the hardware CCs it supports: > - 7.5 which supports hardware CC >=7.5,<8.0 > - 8.0 which supports hardware CC >=8.0,<9.0 except {8.7} > - 8.6 which supports hardware CC >=8.6,<9.0 except {8.7} > - 9.0 which supports hardware CC >=9.0,<10.0 > - 10.0 which supports hardware CC >=10.0,<11.0 except {10.1} > - 12.0 which supports hardware CC >=12.0,<13.0 > Please follow the instructions at https://pytorch.org/get-started/locally/ to install a PyTorch release that supports one of these CUDA versions: 12.6 > ``` > > Did you make any changes to [@webfiltered](https://github.com/webfiltered)'s instructions? Again, I managed to find an older copy of Comfy UI, that works for Pascal based GPUs if that is what you are trying to doi here, such as the 1080 cards. Here is the link. https://www.icloud.com/iclouddrive/03fZ6P6nNqief4fLwmv_9ElBQ#Comfy_UI_Related_Aps_and_Options

bsakaguchi referenced this issue

2026-10-31 22:17:12 +00:00

[PR #363] [MERGED] Installer window #820

Rows
Columns

[GH-ISSUE #1300] Pytorch support for Pascal GPUs #363

Prerequisites

Problem Statement

Proposed Solution

Potential workflow

Impact Level

Alternatives Considered

Additional information

Why this happened

Replace torch from inside the desktop app

Desktop app won't load