Pytorch CUDA error: no kernel image is available for execution on the device on RTX 3090 with cuda 11.1

Question

If I run the following:

import torch
import sys
print('A', sys.version)
print('B', torch.__version__)
print('C', torch.cuda.is_available())
print('D', torch.backends.cudnn.enabled)
device = torch.device('cuda')
print('E', torch.cuda.get_device_properties(device))
print('F', torch.tensor([1.0, 2.0]).cuda())

I get this:

A 3.7.5 (default, Nov  7 2019, 10:50:52) 
[GCC 8.3.0]
B 1.8.0.dev20210115+cu110
C True
D True
E _CudaDeviceProperties(name='GeForce RTX 3090', major=8, minor=6, total_memory=24267MB, multi_processor_count=82)
F 
<stacktrace>
CUDA error: no kernel image is available for execution on the device

More info about my system:

• Nvidia version: NVIDIA-SMI 455.38 Driver Version: 455.38 CUDA Version: 11.1
• python 3.7, Ubuntu 18.04

Answer 1

Found a fix for my problem here: https://github.com/pytorch/pytorch/issues/31285#issuecomment-739139454

pip install --pre torch torchvision -f https://download.pytorch.org/whl/nightly/cu110/torch_nightly.html -U

Then my code snippet gives:

A 3.7.5 (default, Nov  7 2019, 10:50:52) 
[GCC 8.3.0]
B 1.8.0.dev20210115+cu110
C True
D True
E _CudaDeviceProperties(name='GeForce RTX 3090', major=8, minor=6, total_memory=24267MB, multi_processor_count=82)
F tensor([1., 2.], device='cuda:0')

Answer 2

https://pytorch.org/get-started/locally

Use this page to select the configuration of the system. And run the command that is displayed. It works in all kinds of installation problems.