Minimum/maximum CUDA Toolkit requirement for R2022a

Hi,

I'm trying to make sure that the CUDA Toolkit version I have installed is compatible with the version of Matlab that I use, as I've heard that parallel computing can be very sensitive to this.

Looking at the current documentation, there are mention of the compute capability requirement, but nothing about CUDA.

And the older documentation does mention CUDA Toolkit versions, but only up to R2021b.

I am using Matlab R2022a and CUDA Toolkit 11.7, are they compatible ?

Answer 1

[Edric Ellis]

To run code using gpuArray, you do not need to install a CUDA Toolkit. You only need a CUDA driver. You should always* install the latest CUDA driver for your device. (*In the past, there have been rare occasions where the very latest CUDA driver has caused problems - but in general, it is almost always the right thing to install the lastest CUDA driver).

You can compile CUDA MEX files using mexcuda without installing a CUDA Toolkit as (since R2019b) Parallel Computing Toolbox ships the correct version of the CUDA Toolkit to do this.

From this page in the doc:

If you want to generate CUDA® kernel objects from CU code or use GPU Coder™ to compile CUDA compatible source code, libraries, and executables, you must install a CUDA Toolkit. The CUDA Toolkit contains CUDA libraries and tools for compilation.

The precise CUDA Toolkit needed for those cases is mentioned in the release notes. R2022a uses CUDA Toolkit version 11.2.