How to fix Fooocus for RTX 5000 blackwell (windows)

[d3stroyahh]

Sorry if I'm opening a thread but I encountered this issue yesterday and found very old and confusing instructions to make Fooocus work with 5000 series gpus.
Fooocus uses a dedicated embedded python 3.10 with all of the required modules.
You don't need to install torch in your main python, but in the embedded one in fooocus! (I saw many people lost here!)

So basically:

1. Make sure you have Cuda installed, version 12.8 https://developer.nvidia.com/cuda-12-8-0-download-archive
2. now enter the python_embedded folder in foooocus, open a CMD terminal here
3. type python.exe -m pip install --force-reinstall torch==2.7.0+cu128 --extra-index-url https://download.pytorch.org/whl/cu128

Now torch 2.7 officially supports RTX5000 so you don't have to mess with nightly or dev stuff!

Now it will work. If it still does not maybe you just have to install or update torchvision+torchaudio in the same way.

Hope this helps

[Rogala]

You do not need to install CUDA Toolkit 12.8 Downloads, just update the Nvidia drivers to the latest version, go to the python_embedded folder, open the terminal and enter 3 lines, the first one to update the pip, then install the libraries and check their version.

Open the terminal and check the CUDA version with the nvidia-smi command, it should be >=12.8

.\python.exe -m pip install --upgrade pip
.\python.exe -m pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu128
.\python.exe -m pip list

[d3stroyahh]

great! wish I could have had the same clear explanation yesterday..

[postgoatleg]

hi, RTX 5070, CUDA version 12.9, when try to generate an image - i get this error:

RuntimeError: CUDA error: no kernel image is available for execution on the device
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

[Rogala]

Rebuild Torch by opening a terminal in the python_embedded folder, uninstalling your version, and installing a new one.

.\python.exe -m pip uninstall torch torchvision torchaudio
.\python.exe -m pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu128

[DeclanE47]

Rebuild Torch by opening a terminal in the python_embedded folder, uninstalling your version, and installing a new one.

.\python.exe -m pip uninstall torch torchvision torchaudio
.\python.exe -m pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu128

Thank you help its much appreciated 👍

[vaka-ai]

There is a batch file someone made to fix the issues with the 5000 series. i can't remember where I found it but I probably still have it somewhere if anyone needs it. I have a rtx 5070 and it fixed my problems.

Also I'm not sure if this the place for this or allowed but I just started up a YouTube channel showing some AI models I created. I'd love to get some followers. The content is 18+ only.

https://youtu.be/9mthI3s4tOk?si=ewdmZOnyGA94f3RV