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

Hi All

! I’m a new learner of Pytorch, today I try the example of tutorial, the CUDA error happens.

It seems that the opreation move tensor from cpu to GPU is not allowed. Does pytroch not support GT 710? But as shown in the figure, the flag of GPU available is Ture. I am confusing…

The configurations:

Windows 10 64bit,
Anaconda3,
Python 3.7.5,
pytorch 1.3.1 ( py3.7_cuda101_cudnn7_0 )
torchvision 0.4.2 ( py37_cu101 )

And the status of GPU is shown in the pic.

What is the error comes from? Should I adjust somethere to let it work or just need to change an advanced GPU?

Thanks all !

Bree

The 1.3.1 binaries support GPUs with compute capability >= 3.7, while your GPU seems to have 3.5.
You could build from source as described here.

Thanks for your reply!

I follow the instructions to build from source, and when “python setup.py install” meet cmake error as shown in the figure:

It seems that the C++ compiler cannot to compile a sample test program. But I installed Visual Studio 2019, the compiler should be the latest, why is the error happend? The conda list is also shown at the end of the fig. How to fix this error?

Thank you.

You may need to update cmake to a newer version. Please also remember to do python setup.py install --cmake next time if you build again.

Thanks for your reply!

I follow the instructions to build from source, and when “python setup.py install” meet cmake error as shown in the figure:

It seems that the C++ compiler cannot to compile a sample test program. But I installed Visual Studio 2019, the compiler should be the latest, why is the error happend? The conda list is also shown at the end of the fig. How to fix this error?

Thank you.

Thank you my friend.

I have conda update cmake to the lastest, but it still got the wrong answer.

https://cmake.org/download/ The latest version should be 3.16.1. It is also available in the conda-forge channel. https://anaconda.org/conda-forge/cmake

Sorry I forgot to upload the screenshot after I followed your first advice. But I do update cmake to version 3.16.1, and it does not work…

What about doing the following?

set USE_NINJA=OFF
python setup.py install --cmake