RuntimeError: CUDA error: no kernel image is available for execution on the device

Graphics / Linux Linux
1

Hallo, I’m trying to set up an new virtural environment and run the model, but the following error shows up: RuntimeError: CUDA error: no kernel image is available for execution on the device

The system I am using is:

  • Ubuntu 20.04
  • laptop GeForce RTX 3070
  • Driver Version: 510.73.05
  • torch=1.2.0 in virtualenv environment
  • install cuda 10.0 with .runfile, also check with nvcc -V

A simple test has been launched:

import torch

a = torch.randn(1,2,3).cuda()
print(a)

It works but too long! (about 2-3min) as:

tensor([[[ 0.9284, -0.6863,  1.5324],
         [-0.6943, -0.7309,  0.4542]]], device='cuda:0')

When I run the network model, the program also takes long and finally raise:
RuntimeError: CUDA error: no kernel image is available for execution on the device

Could you please give me some advice? Thanks in advance!

2

The versions of CUDA and PyTorch follow the model requirements.

3

Cuda 10 is incompatible to the 3070, you need cuda 11.

1 Like