The MathWorks' Silvina Grad-Freilich talks high-performance computing

Rick Nelson, Editor-in-Chief -- EDN, May 27, 2010

What does The MathWorks offer that can help designers take advantage of HPC?

A: We have a set of tools called the Parallel Computing Toolbox, and these tools allow you to solve larger problems by using additional compute capacity. That additional compute capacity can be anything from the multiple cores in your desktop machine to clusters or grids. A second toolset is Matlab Distributed Computing Server, which allows users to scale the applications they develop to run on a cluster.

How can Matlab and Simulink users employ these tools?

A: Matlab users usually write their own code, but one of the premises is that they are domain experts more than programmers. One of the main reasons they came to Matlab is to basically have a development environment where they can rapidly develop a prototype. One of the things that these parallel-computing tools allows them to do is stay in the same environment.

What other support do you offer for parallel computing?

A: We have integrated other toolboxes with the Parallel Computing Toolbox. In the Statistics Toolbox and Optimization Toolbox, for example, there are already some functions we have parallelized. So if you write an application based on those functions, you don't need to change your application at all to be able to use all the cores in your machine or scale your application up to a cluster.

In addition to that feature, we provide APIs [application-programming interfaces] so that you as a Matlab user can also parallelize your own applications. For example, let's suppose that you have an application such as a parameter sweep or a Monte Carlo simulation that tests Matlab code over a set of parameters. In Matlab, that would be expressed by a for loop. To parallelize the application, you just change the for loop to a parfor loop.

And what about Simulink?

A: Something very similar happens to Simulink. For example, the Simulink Design Optimization tool also works with the Parallel Computing Toolbox. With parallel computing integrated within Design Optimization, you can make use of the different cores in your machine or a cluster to run more simulations.

IDC predicts that x86 processors will dominate but GPUs (graphics-processing units) will gain traction. Does The MathWorks support GPUs?

A: Not yet, but we have a beta program running right now that started at the beginning of this year. Basically, we are inviting Matlab users to test our solution for GPUs. The program has been amazingly popular within our customer base.

Were there any surprising predictions raised in the IDC presentation?

A: Not surprising, but interesting. One prediction is about how the challenge of highly parallel programming will increase, and we as programmers have seen all the time that programming a multicore machine is significantly harder than programming for a single-core machine, but what we are trying to do with our tools is to shield users from that complexity.

What else did you take note of in the predictions?

A: One of the predictions is that the HPC market will resume growth in mid-2010. That prediction is great to see. One thing we have said is that, during the recession and even now, our tools have been very successful. Even with all the changes in the economy and the recession and layoffs, people are looking for more and more ways to be more productive, and running our tools on high-performance computers is all about productivity.