How do I use multiple GPU for GAN

In the example metioned on MATLAB "Train Generative Adversarial Network (GAN) - MATLAB & Simulink (mathworks.com)" how/where should the code be changed so as to use multiple GPU's?

Though, "auto" is used, all the GPU's are not being used by default. I have 4 gpu's and want to use them all.

[Shuaibin WAN]

Hi Shaw,

I also encounter this problem. Did you have any solution now?

Many thanks!

Answer 1

[Antti]

Hi! You should change 'ExecutionEnvironment' option to 'multi-gpu'. More info here. Before doing that, you might want to check if your GPU's are detected b

>> numGPU = gpuDeviceCount("available")

you don't get 4 as a result, then your GPU's are not supported by MATLAB, or there's a driver issue. Please accept my answer formally if this worked for you.

[Shaw]

Hey, I have already tried using that. It throws me an error:

"The input, "multi-gpu" did not match any of the valid values."

Valid values being, 'auto' ,'gpu' ,'cpu'

And yes, I got 4 as a result for my GPUcount

Answer 2

[Antti]

It appears that when using custom training loops (as in the example), "multi-gpu" option is not supported. However, you can still take advantage multiple GPUs, by launching parallel MATLAB workers, where each worker will use a GPU of its own. See this example: https://se.mathworks.com/help/deeplearning/ug/train-network-in-parallel-with-custom-training-loop.html. Please formally accept my answer if this solves your problem.