Did I buy a 5080 for nothing?

[serget2]

Your question

Oh you should upgrade your card, your card is obsolete you only have 12 GB vram, so I am patiently waiting for a gpu to upgrade, patiently waiting for prices to drop (only to have that oaf of a president from dumbfckistan put tarrifs on everyone) to find out that, Hey I don't have money for a 5090 but for half the price I could buy a 5080 and have just enough Vram to play with the latest and greatest, only to find out ComfyUI does not support the 50XX series? I needed to download a spezjul version that works under windows with the correct Cuda? I wanted to do Framepack, Hunyuan and WAN 2.1 videos but that sht doesn't install, can't find the nodes, can't run any video workflow? Tell me: What did I buy a 5080 RTX for?

Logs

Other

No response

[brknsoul]

ComfyUI does not support the 50XX series

It certainly does.

How to Update Torch for 5000 series GPUs

portable zip

open cmd in comfyui_windows_portable folder and run; python_embeded\python -m pip install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

desktop app

open bottom panel > terminal tab, run; pip install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

[samsonsite1]

I think RTX 50xx series need at least CUDA 12.8 installed for optimal support. So, you should start by downloading and installing CUDA Toolkit 12.8 runtimes and binaries.

[pocketpoetry]

you may find this video useful to sucessfully install Comfyui for RTX 50 series, with 12.8 built in https://youtu.be/7JJFJr8doPw?si=9er5r-bTPaN7oGYI

Alternatively for better control of what you are installing, can try manual install of the Pytorch 2.7 with CUDA 12.8 built https://youtu.be/jz7ueLmrvY4?si=oujGB7QPOmkIKMzI

50 series is supported, just that it runs on a newer Pytorch version which is an important element in Comfyui, its catching up slowly to be the norm, you are certainly on the right path to get a 50 series card, just that in the world of open source AI, having the money to buy a better GPU means we are given the chance to work harder to solve problems before we get the cake, instead of just paying and we get the cake.. haha.. patience and you will understand it soon enough =)

[brknsoul]

Just note: release 0.3.30 now comes with 2.7.0+cu128 by default.

[woct0rdho]

@samsonsite1 To clarify the myth, you don't need to manually install the full 'CUDA toolkit' that takes many GB. The torch wheel already bundles the necessary CUDA libraries.

[samsonsite1]

I download it out of habit, as I have other tools installed that aren't bundled with CUDA runtime binaries. And it only takes up 3 GBs. That's nothing compared to some of the diffusion models I have to download.

[github-actions[bot]]

This issue is being marked stale because it has not had any activity for 30 days. Reply below within 7 days if your issue still isn't solved, and it will be left open. Otherwise, the issue will be closed automatically.