It doesn't work with the RTX5080.

[mizossyo]

NVIDIA GeForce RTX 5080 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 5080 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

This message appears when starting up and generation is no longer possible. In other words, since the new GPU RTX5080 does not support CUDA121, I think it is necessary to rebuild the venv with a new version of CUDA. (For example, CUDA126?)

However, when trying to build a venv from user.bat, the file is forcibly built for the CUDA121 environment, which is inconsistent with CUDA126 and makes it impossible to even install. Can someone please tell me how to solve this?

It may be a problem with the RTX5080 driver...

[allerias]

Same issue with RTX5090

[dan4ik94]

Nvidia spamming that AI AI AI AI everywhere but they cant communicate with real AI programmists to release their card and drivers all compatible with all environements. This should be highlighted everywhere, and people responsible for the lack of support on day 3 of release of that cards should be fired! Or nvidia to just drop so low on stock values that would send them forever to polish AMD shoes. Not to mention that they released DLSS 4 but only people that know how to correctly install them to games can enjoy the transformer model of dlss. This is really a circus and literally worst release ever of their cards. I was lucky to grab 5080 but i really want right now to sell it so i can forget about headache configuring of that card gave me already.

nobody's gonna communicate anything, there are too many caveats, everyone with their own affairs and schedule. Just wait a week or two, new pytorch pip should drop soon.

[Panchovix]

if using linux you can use the cu128 nightly builds, if on windows you have to use the wheels from nvidia on HF

[mizossyo]

Thank you. I downloaded and installed the wheel from nvidia and changed the version in requirements_versions, but it didn't work. Stupid me!

[rubentorresbonet]

same issue with 5090

[yeonpyo3300]

same with 5080 comfy ui

[Slider-Override]

Is there any workaround ? I am not a expert on this but can we maybe intigrate this into forge? https://huggingface.co/w-e-w/torch-2.6.0-cu128.nv

If yes, whats the steps ?

[zlo899nv]

Hi all! Update from our NVIDIA team.

To use PyTorch for Linux x86_64 and Linux SBSA on NVIDIA 5080, 5090 Blackwell RTX GPUs use the latest nightly builds, or the command below.

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

Stay tuned for further updates

Thanks!

[Slider-Override]

Update from our NVIDIA team

To use PyTorch for Linux x86_64 on NVIDIA 5080, 5090 Blackwell RTX GPUs use the latest nightly builds, or the command below. pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

for Linux SBSA pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu128

torchvision and torchaudio for SBSA coming soon

Thanks!

Thanks, but no solution for Windows currently ?

[rubentorresbonet]

Update from our NVIDIA team To use PyTorch for Linux x86_64 on NVIDIA 5080, 5090 Blackwell RTX GPUs use the latest nightly builds, or the command below. pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128 for Linux SBSA pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu128 torchvision and torchaudio for SBSA coming soon Thanks!

Thanks, but no solution for Windows currently ?

Have a look here, they talk about windows and docker container: https://github.com/comfyanonymous/ComfyUI/discussions/6643

[zebulon2]

Hi all! Update from our NVIDIA team.

To use PyTorch for Linux x86_64 and Linux SBSA on NVIDIA 5080, 5090 Blackwell RTX GPUs use the latest nightly builds, or the command below.

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

Stay tuned for further updates

Thanks!

Thank you. This worked with Fooocus, but not with latest WebUI-Forge: RuntimeError: Failed to import diffusers.pipelines.pipeline_utils because of the following error (look up to see its traceback): Failed to import diffusers.models.autoencoders.autoencoder_kl because of the following error (look up to see its traceback): No module named 'triton._C.libtriton.triton'; 'triton._C.libtriton' is not a package

Have I made a mistake?

EDIT: I was able to fix some errors after updating transformers and triton, but still get this error: No module named 'triton.ops'

[shiraishi-png]

Hi all! Update from our NVIDIA team. To use PyTorch for Linux x86_64 and Linux SBSA on NVIDIA 5080, 5090 Blackwell RTX GPUs use the latest nightly builds, or the command below. pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128 Stay tuned for further updates Thanks!

Thank you. This worked with Fooocus, but not with latest WebUI-Forge: RuntimeError: Failed to import diffusers.pipelines.pipeline_utils because of the following error (look up to see its traceback): Failed to import diffusers.models.autoencoders.autoencoder_kl because of the following error (look up to see its traceback): No module named 'triton._C.libtriton.triton'; 'triton._C.libtriton' is not a package

Have I made a mistake?

EDIT: I was able to fix some errors after updating transformers and triton, but still get this error: No module named 'triton.ops'

I am getting the same error, does anyone have any fixes? EDIT: fixed it by downgrading triton version to 3.1.0

Another issue I have run into: WARNING:bitsandbytes.cextension:Could not find the bitsandbytes CUDA binary at PosixPath('/home/victor/stable-diffusion-webui-forge/venv/lib/python3.12/site-packages/bitsandbytes/libbitsandbytes_cuda128.so') WARNING:bitsandbytes.cextension:The installed version of bitsandbytes was compiled without GPU support. 8-bit optimizers, 8-bit multiplication, and GPU quantization are unavailable.

Any ideas?

[Natural-Warp]

looking forward to the windows solution indeed ( SD ForgeUI on 5080 here )

[magmirant0011]

looking forward to the windows solution indeed ( SD ForgeUI on 5080 here )

You need to reinstall torch and torchvision. Enter anaconda prompt, get inside venv of the Forge, uninstall existing torch+torchvision, uninstall xformers if you use it on Forge, download torch+torchvision that is compatible with Blackwell GPU and install it to venv with the anaconda prompt. I did that and now my 5080 can do anything in A111 but in Forge is almost same.

[shiraishi-png]

I got it working, but when using ADetailer more than once I get this error:

ValueError: Weights only load failed. This file can still be loaded, to do so you have two options, do those  │
    │ steps only if you trust the source of the checkpoint.                                                         │
    │         (1) In PyTorch 2.6, we changed the default value of the `weights_only` argument in `torch.load` from     │
    │ `False` to `True`. Re-running `torch.load` with `weights_only` set to `False` will likely succeed, but it can    │
    │ result in arbitrary code execution. Do it only if you got the file from a trusted source.                        │
    │         (2) Alternatively, to load with `weights_only=True` please check the recommended steps in the following  │
    │ error message.                                                                                                   │
    │         WeightsUnpickler error: Unsupported global: GLOBAL ultralytics.nn.tasks.DetectionModel was not an        │
    │ allowed global by default. Please use `torch.serialization.add_safe_globals([DetectionModel])` or the            │
    │ `torch.serialization.safe_globals([DetectionModel])` context manager to allowlist this global if you trust this  │
    │ class/function.

~~The file then becomes corrupted. The only "solution" is to use ADetailer once, then restart the whole server, then use it again, repeat... Less than ideal, but it works. Does anyone know how to fix this?~~

Edit: fixed it by adding:

import os
os.environ['TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD'] = '1'

to innit.py in torch lib located in env and to launch.py on main folder, but I have no idea what I'm doing so yeah

[licongy]

looking forward to the windows solution indeed ( SD ForgeUI on 5080 here )

You need to reinstall torch and torchvision. Enter anaconda prompt, get inside venv of the Forge, uninstall existing torch+torchvision, uninstall xformers if you use it on Forge, download torch+torchvision that is compatible with Blackwell GPU and install it to venv with the anaconda prompt. I did that and now my 5080 can do anything in A111 but in Forge is almost same.

What's your platform? Linux or Windows?

[magmirant0011]

looking forward to the windows solution indeed ( SD ForgeUI on 5080 here )

You need to reinstall torch and torchvision. Enter anaconda prompt, get inside venv of the Forge, uninstall existing torch+torchvision, uninstall xformers if you use it on Forge, download torch+torchvision that is compatible with Blackwell GPU and install it to venv with the anaconda prompt. I did that and now my 5080 can do anything in A111 but in Forge is almost same.

What's your platform? Linux or Windows?

Win10 Ryzen platform.

[allerias]

looking forward to the windows solution indeed ( SD ForgeUI on 5080 here )

You need to reinstall torch and torchvision. Enter anaconda prompt, get inside venv of the Forge, uninstall existing torch+torchvision, uninstall xformers if you use it on Forge, download torch+torchvision that is compatible with Blackwell GPU and install it to venv with the anaconda prompt. I did that and now my 5080 can do anything in A111 but in Forge is almost same.

Can you give some examples of the above commands ? I don't find how to uninstall torch from the venv of Forge.

[shiraishi-png]

looking forward to the windows solution indeed ( SD ForgeUI on 5080 here )

You need to reinstall torch and torchvision. Enter anaconda prompt, get inside venv of the Forge, uninstall existing torch+torchvision, uninstall xformers if you use it on Forge, download torch+torchvision that is compatible with Blackwell GPU and install it to venv with the anaconda prompt. I did that and now my 5080 can do anything in A111 but in Forge is almost same.

Can you give some examples of the above commands ? I don't find how to uninstall torch from the venv of Forge.

Are you on Windows or Linux? If on Windows, go to forge folder, then venv/scripts folder, open cmd and type: ./activate now type pip uninstall torch

if on Linux go to webui forge directory, open cmd and type: source venv/bin/activate now type pip uninstall torch if on Linux you can install the nightly cu128 build that is linked in this thread while on venv and it should work. I haven't managed to get it working on Windows though, so I'm not sure what to do there.

[licongy]

torch nightly version for cuda 128/Windows has just been released. I'm guessing the RTX 5080 can run sd webui forge now.

[Natural-Warp]

torch nightly version for cuda 128/Windows has just been released. I'm guessing the RTX 5080 can run sd webui forge now.

any guidance or tips on how-to ??

[xiao3612736]

NVIDIA GeForce RTX 5080 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 5080 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

Operating system: win11 x64 I have tried cu118, cu121, cu126, and also tried changing the RTX5080 graphics card driver, but the problem still cannot be solved. God, save me!!!

[shiraishi-png]

NVIDIA GeForce RTX 5080 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 5080 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

Operating system: win11 x64 I have tried cu118, cu121, cu126, and also tried changing the RTX5080 graphics card driver, but the problem still cannot be solved. God, save me!!!

A clean install of ReForge worked for me flawlessly on windows, it automatically downloads the nightly cu128 version for windows.

[allerias]

I can run forge again after do the following easy steps:

• Perform a fresh install of ReForge (https://github.com/Panchovix/stable-diffusion-webui-reForge)
• Remove Forge venv folder
• Copy the venv folder of Reforge into the Forge directory

[Natural-Warp]

A clean install of ReForge worked for me flawlessly on windows, it automatically downloads the nightly cu128 version for windows.

Can you post a link and step guide please ?

[shiraishi-png]

A clean install of ReForge worked for me flawlessly on windows, it automatically downloads the nightly cu128 version for windows.

Can you post a link and step guide please ?

https://github.com/Panchovix/stable-diffusion-webui-reForge

Follow the install instructions (Clean install), that's all I had to do for it to work

[allerias]

Reforge works after a fresh install but it's not Forge

[shiraishi-png]

Reforge works after a fresh install but it's not Forge

Correct. Therefore, if your solution of copying the venv from ReForge into Forge works, then in my opinion it still applies to help @Natural-Warp to install ReForge, since they will need to be able to install it to apply the solution.

[guiteubeuh]

I can run forge again after do the following easy steps:

* Perform a fresh install of ReForge (https://github.com/Panchovix/stable-diffusion-webui-reForge)

* Remove Forge venv folder

* Copy the venv folder of Reforge into the Forge directory

Where is the 'venv' FORGE folder located? I can't find it anywhere. (Windows)

[killua132445]

I can run forge again after do the following easy steps:

* Perform a fresh install of ReForge (https://github.com/Panchovix/stable-diffusion-webui-reForge)

* Remove Forge venv folder

* Copy the venv folder of Reforge into the Forge directory

Where is the 'venv' FORGE folder located? I can't find it anywhere. (Windows)

For me it's in (stable-diffusion-forge main folder)/webui/venv. Let me know if it works for you since I got to run the ssd reforge but i'm getting the same cuda kernel error on forge.