MATLAB: R2018a can’t access gpu card

gpur2018a

Dear MathWorks Support Team,
I have installed matlab R1028a on an HPC cluster. When I try to access one of the gpu nodes I get the following error message:
gpuDevice()
Error using gpuDevice (line 26)
There is a problem with the graphics driver or with this GPU device. Be sure that you have a <a href="matlab:web('http://www.mathworks.com/gpudevice','-browser')">supported GPU</a> and that the <a
href="matlab:web('http://www.mathworks.com/gpudriver','-browser')">latest driver</a> is installed.
Caused by:
    The graphics driver was found, but it is too old. The graphics driver on your system supports CUDA version 8. The required CUDA version is 9 or greater.
The nvidia drivers installed in this particular node are:
There is no problem with earlier releases of matlab installed in the cluster, i.e. R2017a.
Other gpu nodes in the cluster have a recent version of the nvidia nodes, namely 396.26. In this case matlab R2018a can run computations on gpu cards.
Is matlab R2018a incompatible with old versions of the nvidia drivers installed for this specific gpu card?
Thanks

Best Answer

Is matlab R2018a incompatible with old versions of the nvidia drivers installed for this specific gpu card?
Yes. The 396 series are CUDA 9, but the 367 series are CUDA 8, and MATLAB R2018a needs CUDA 9
https://www.mathworks.com/help/distcomp/release-notes-R2018a.html
"The parallel computing products are now using CUDA® Toolkit version 9.0. To compile CUDA code for CUDAKernel or CUDA MEX-files, you must use toolkit version 9.0."
The corresponding entry for R2018b (newest) is
"The parallel computing products are now using CUDA® Toolkit version 9.1. To compile CUDA code for CUDAKernel or CUDA MEX-files, you must use toolkit version 9.1."
Related Question