Matlab 2020a-AWS cluster

Hi , am just set up a cluster on AWS with the setting:

MATLAB Version:R2020a

Shared State:Personal

Auto-Manage Cluster Access:Yes

Worker Machine Type:Double Precision GPU (p3.8xlarge, 16 core, 4 GPU)

Workers per Machine:16

Use a dedicated headnode:Yes

Headnode Machine Type:Standard (m5.xlarge, 2 core)

Machines Requested:2

Machines Available:0

Allow cluster to auto-resize:No

Access to MATLAB Drive:Yes

Initial Worker Count:16

and got it running online, however, when I set this cluster as a defult for my matlab, when I use gpuDevice, I think I am only finding the gpu on my own computer, not the cluster. What did I do wrong or is there a mis-setting.

Answer 1

[Raymond Norris]

Hi Kevin,

Start a parallel pool and then run your GPU code from within a parallel construct. For example:

% Time x = A\b on CPU and GPU

% Start parallel pool on AWS
parpool(32);

spmd
    sz = 2^14;
    
    t0 = tic;
    A = rand(sz);
    b = rand(sz,1);
    x = A\b;
    cpu_t = toc(t0)
    
    t1 = tic;
    gA = gpuArrya.rand(sz);
    gb = gpuArray.rand(sz,1);
    gx = gA\gb;
    x2 = gather(gx);
    gpu_t = toc(t1)
end

Keep in mind that the GPU might need to warm up the first time through, so you might need to run it again.

Raymond