ValueError: Unknown CUDA arch (8.6) or GPU not supported

[TDLord0]

windows10 python 3.6.13 cuda 10.1 pytorch 1.4.0

gpu rtx3080

when (python setup.py build develop)- the Cuda 10.1 is not supported my GPU so I try to use Cuda 11.x but also has a lot of error... plz help

[samymdihi]

I faced the same issue. You need to download the latest version of Cuda : Cuda 11.3. Then use the commands conda install -c anaconda cudatoolkit and conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch -c nvidia when creating your virtaul env. It will install cudatoolkit 11 and the corresponding pytorch and torchvision versions.

Also be sure to use the branch that supports pytorch > 1.5. To do so, use the commands :

git clone https://github.com/MVIG-SJTU/AlphaPose.git cd AlphaPose git pull origin pull/592/head

You can then finish the installation

[assafge]

conda didn't worked for me either, pip did the job: pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 -f https://download.pytorch.org/whl/torch_stable.html

and then follow Alphapose installation instructions.

[RolandCita]

Might be a bit late, but what did it for me, was installing pytorch 1.10.1. Newer versions threw THCuda errors, older ones CHECK_CUDA errors.

conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge

GPU used is RTX 3090 with cuda 11.7 system-wide