Building Pytorch from source pytorch binaries

1

i am facing the problem
GeForce RTX 3090 with CUDA capability sm_86 is not compatible with the current PyTorch installation

" i also dont know about binaries and what is mean by " building Pytorch from source ? and what is other way to build pytorch which can work me and can solve this problem "

2

Double post from here with the answer.
Install the CUDA11 binaries using the install instructions from the website and your 3090 should work.

3

@ptrblck , my cuda version is : cuda_11.5 (Build cuda_11.5.r11.5/compiler.30672275_0
) , torch version is: 1.10.2+cu102. However, on this link, there is no option for Cuda version 11.5. When I use Cuda version 11.6, I still face the issue:

UserWarning: 
NVIDIA GeForce RTX 3090 with CUDA capability sm_86 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_70.
If you want to use the NVIDIA GeForce RTX 3090 GPU with PyTorch

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.

Can you please help me solve this issue?

4

That’s the wrong binary as it’s using CUDA 10.2 while you need CUDA 11.x for your 3090.

Select 11.3 or 11.6.

Make sure you have only the one right binary installed and uninstall the old ones.

1 Like
5

@ptrblck thank you, it worked.