UserWarning: NVIDIA GeForce RTX 5070 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.

[fireYtail]

Expected Behavior

There shouldn't be any errors on a fresh install, it should work regardless of the GPU. Alternatively, it should automatically detect the incompatibility and try to fix it.

The user shouldn't have to manually reinstall. This workaround is far from an acceptable solution, A lot of people don't know how to fix this issue (the manual reinstall). You shouldn't expect the average ComfyUI user to know how to fix it entirely by themselves.

Actual Behavior

On a fresh install (or an older install), when using a 5070 ti the following error comes up:

A:\Sin Sincronización\Chrome\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
Checkpoint files will always be loaded safely.
A:\Sin Sincronización\Chrome\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py:235: UserWarning:
NVIDIA GeForce RTX 5070 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5070 Ti GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

There is a way to fix this, but the user has to follow these specific steps manually:

1. Navigate to python_embedded\ directory and open a new command prompt in this location.
2. Run python -m pip uninstall torch torchvision
3. Run python -m pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu128 and wait for a very long download of over 3 GB.

Steps to Reproduce

With a 5070 ti as your GPU, make a fresh install of ComfyUI and run it (or run an already existing installation)

Debug Logs

A:\Sin Sincronización\Chrome\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
Checkpoint files will always be loaded safely.
A:\Sin Sincronización\Chrome\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py:235: UserWarning:
NVIDIA GeForce RTX 5070 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5070 Ti GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

Other

I upgraded from a 1080 ti to a 5070 ti, but this isn't relevant since it happens on a fresh install (latest release) too.

[LukeG89]

It's "normal" behaviour, as you can see from the NVIDIA 50xx discussion:

• https://github.com/comfyanonymous/ComfyUI/discussions/6643

In the discussion, there is the download link to a ComfyUI portable with nightly Pytorch already installed.

Stable Pytorch is currently incompatible with 50 series GPUs, but I guess it's just a matter of time.

[fireYtail]

It's "normal" behaviour, as you can see from the NVIDIA 50xx discussion:

* [Nvidia 50 Series (Blackwell) support thread: How to get ComfyUI running on your new 50 series GPU. #6643](https://github.com/comfyanonymous/ComfyUI/discussions/6643)

In the discussion, there is the download link to a ComfyUI portable with nightly Pytorch already installed.

Stable Pytorch is currently incompatible with 50 series GPUs, but I guess it's just a matter of time.

The link is on the issue page, not on the main page of the GitHub repo. This is not how to do things properly. People won't go into every issue of every GitHub repo just to get working versions. Expecting users to do this is unrealistic.

Is it so tremendously difficult to just add a small note to the main page about it? For God's sake.

[LukeG89]

Actually, there is the link in the readme to the discussion I linked: https://github.com/comfyanonymous/ComfyUI?tab=readme-ov-file#windows-portable

There is also a note in the NVIDIA section about nightly pytorch for 50xx GPUs: https://github.com/comfyanonymous/ComfyUI?tab=readme-ov-file#nvidia

[canglangzhishuiz]

I solved this issue on Windows 11 with an RTX 5070 Ti and WSL2.

First, install CUDA 12.8.
Then, install PyTorch using:

pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

Finally, clone the PyTorch3D source code and install it:

git clone https://github.com/facebookresearch/pytorch3d.git
cd pytorch3d
pip install -e .

[chaxheri]

Same problem when use 5070Ti，is any updating for this issue?

[calvinwonghk]

Thanks for tips! Windows desktop installer version Just go python path is : C:\Users<user>\Documents\ComfyUI.venv\Scripts

pip uninstall torch pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

now is working fine, thanks for tips!

[spha-code]

Thanks for tips! Windows desktop installer version Just go python path is : C:\Users\Documents\ComfyUI.venv\Scripts

pip uninstall torch pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

now is working fine, thanks for tips!

This worked for me!

[akaghzi]

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

worked for me, i am using ubuntu 24

[aotianlong]

Thanks for tips! Windows desktop installer version Just go python path is : C:\Users\Documents\ComfyUI.venv\Scripts

pip uninstall torch pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

now is working fine, thanks for tips!

worked for me , thank you.