{"id":603,"date":"2022-11-07T10:08:39","date_gmt":"2022-11-07T15:08:39","guid":{"rendered":"https:\/\/blogs.mathworks.com\/matlab\/?p=603"},"modified":"2023-10-25T07:37:56","modified_gmt":"2023-10-25T11:37:56","slug":"matlabs-high-performance-computing-hpc-and-big-data-datatypes","status":"publish","type":"post","link":"https:\/\/blogs.mathworks.com\/matlab\/2022\/11\/07\/matlabs-high-performance-computing-hpc-and-big-data-datatypes\/","title":{"rendered":"MATLAB’s High Performance Computing (HPC) and ‘Big Data’ datatypes"},"content":{"rendered":"
Table of Contents<\/span><\/div>
gpuArrays - GPU programming made easy\r\n<\/span><\/a>dlarray - Specialised objects for Deep Learning training\r\n<\/span><\/a>Tall arrays - For when you've got more rows than memory\r\n<\/span><\/a>Distributed arrays - Spread your arrays across multiple nodes of a HPC cluster\r\n<\/span><\/a>Codistributed arrays - Advanced moves for distributed array users<\/span><\/a><\/div><\/div>
I'm currently preparing for <\/span>SC22<\/span><\/a>, my first supercomputing conference as a MathWorker. While making plans with old friends from other High Performance Computing organisations who don't know MATLAB so well, I'm often asked 'HPC in MATLAB....What else is there besides <\/span>parfor<\/span>?'<\/span><\/div>
parfor<\/span><\/a> <\/span>is great, allowing users to go from a serial loop to a parallel one with the addition of only 3 letters, but it's just the beginning of the possibilities for parallel and scalable programming in MATLAB. <\/span><\/div>
Today, I wanted to give a brief survey of what I think of as MATLAB's HPC data-types which allow users to make use of GPUs, distributed memory and work with out-of-core data.<\/span><\/div>

gpuArrays - GPU programming made easy<\/span><\/h3>
GPUs<\/span><\/a>, or Graphics Processing Units, can drastically accelerate certain types of computation. MATLAB currently only supports NVIDIA GPUs which are traditionally programmed using a language called <\/span>CUDA<\/span><\/a>. You can code in CUDA if you want to, and interface to MATLAB using <\/span>CUDA mex functions<\/span><\/a>....but there's an easier way.<\/span><\/div>
Say you have an array, <\/span>X<\/span><\/div>
X = [1,2,3];<\/span><\/span><\/div><\/div><\/div>
transfer this to the GPU with <\/span><\/div>
G = gpuArray(X);<\/span><\/span><\/div><\/div><\/div>
Any calculation you subsequently do on the <\/span>gpuarray<\/span> <\/span>G<\/span> will be automatically performed on the GPU.<\/span><\/div>
Gsq = G.^2; <\/span>% Performed on the GPU<\/span><\/span><\/div><\/div>
Gsin = sin(G); <\/span>% Performed on the GPU<\/span><\/span><\/div><\/div><\/div>
As with the other 'HPC datatypes', you use <\/span>gather<\/span> to transfer a result back to a 'normal' MATLAB matrix<\/span><\/div>
Xsq = gather(Gsq);<\/span><\/span><\/div><\/div>
Xsin = gather(Gsin);<\/span><\/span><\/div><\/div><\/div>
gpuArrays require <\/span>Parallel Computing Toolbox<\/span><\/a>. <\/span><\/div>