The HPC Results Are In

By Steve Robbins

Recently we surveyed you on your interest and use of high-performance computer (HPC) systems in engineering applications. I would like to thank the 651 subscribers who took the time to fill out an extensive survey. You let us know how HPC is changing the way you work, where you turn to get the information you need to evaluate HPC, and what would justify your purchase of an HPC cluster. The results will help us provide you with the kind of content you are looking for, as well as to understand how important different channels of information are to making the decision that could cost you anywhere from a few thousand dollars to millions.

You Said It
Interestingly, 59% of respondents either have an HPC cluster, have evaluated purchasing one, or are in the process of evaluating the purchase of a cluster computer. Not surprising, most of the installed base of HPC clusters today are in mid-to-large size companies, while about half of small-to-mid sized companies are currently evaluating the purchase of their first HPC system.

So, where do you go to find information that will help with the decision on acquiring an HPC system? Ninety-eight percent of those considering a purchase said that independent published information was very or somewhat important. Hardware vendors came in second at 93%. I found it interesting that only 15% used another engineering company as a source of information on the purchase of HPC systems.

Of those who have been involved in the purchase of an HPC cluster, performance was the key factor in the purchase, followed by support, with price lagging substantially. Just as important: In-person training from their CAE vendor would have helped respondents in their purchase decision process.

Tracking Trends
For four years now, DE has been surveying our subscribers on their use and views of high-performance computing. The trend of using these enabling technologies to perform analysis and simulations has increased dramatically as the costs have gone down and the performance has increased. While we did not directly ask about ease of use, clusters have become more accessible to the small- to mid-size engineering companies that need them most.

DE’s recent coverage of how Parker Aerospace used a workstation cluster to speed up throughput on CFD simulations demonstrates that most engineering teams have the compute resources for some level of multi-core system. These systems can be implemented without a large IT department and used by any engineering team with the available cores in their workstations. With the advent of multi-core processors, a workstation can have 12 or more cores available.

Building a dedicated cluster is much easier than even a few years ago. Depending on the application, a reasonably scaled cluster can be achieved without the problems inherent to assembly and testing that were once common. See our continuing coverage of HPC clusters on page 16.

Performance ROI
So what is the next question? How much performance do I need? This is where the software vendor participation becomes crucial.

I know management and the financial guys find return on investment hard to quantify, but I’ve been at DE a long time. For the first five years of this magazine, the average engineer—those of you not working in automotive or aerospace—had to justify the purchase of very expensive software. Whether finite element analysis, computational fluid dynamics, or some other analysis application, you were questioned on why you needed it, what it would do that wasn’t already being done, and how it would create a ROI. Those same questions will be asked of HPC. The answers are obvious: When justifying the purchase of a high-performance cluster, it’s not just about doing things faster, it’s about doing them better. Being able to run more simulations creates a better design.

HPC is here to stay. What usually happens, and this is what happened in the Parker Aerospace story, is once implemented, the first cluster shows enough benefits to justify another one. This is why 10% of our subscribers are already considering the purchase of more HPC systems. It might be a small cluster on the department level, or a remote one that will still be faster than waiting for large amounts of data to return from the main office.

Of course, not everyone needs a high-performance computing system, but for most engineers running simulation and analysis jobs on their workstation, I see one in your future.

I’m off to Supercomputing 2011 this month. I’ll report back on what’s new.

Steve Robbins is the CEO of Level 5 Communications and executive editor of DE. Send comments about this subject to [email protected].