I have installed CUDA 8.0.61 and I run python on Ubuntu 16.04.3.
The code is available on-line so that it should not be wrong.
However, I got the error message below:
RuntimeError: ‘nvcc’ command returns non-zero exit status.
command: [‘nvcc’, ‘–cubin’, ‘-arch’, ‘sm_61’, ‘/tmp/tmpsiw3X0/kern.cu’]
nvcc fatal : Value ‘sm_61’ is not defined for option ‘gpu-architecture’
Can anyone help me to solve this problem?
I suspect you have an older version of CUDA installed on your machine somewhere, and python is picking up that older version.
Find all instances of
nvcc on your machine:
sudo find / -name nvcc
Make sure you have set the PATH variable correctly for your CUDA 8.0.61 install. (Compare the path to your 8.0.61 nvcc from the output above, with echo $PATH)
Thanks for your advice.
There are cuda-7.0, cuda-7.5 and cuda-8.0 in my machine.
Do I have to uninstall the older version?
I’m new to linux.
Can you teach me how to make sure the PATH variable? Thank you very much.
that is installed on your machine
especially the section titled “Post Installation Actions”
and the nvcc here:
may be troublesome. You’ll need to make sure that the path to /usr/local/cuda-8.0/bin appears before that path in your PATH variable.