-by Richard Fichera

I recently had an opportunity to take a look at the latest Top 500 supercomputer list. Like a lot of people, I was awed by the performance and the churn at the top of the list. Working for a company that makes lots of x86 blade servers, I was pleased to note that several of the top 10 are now clusters of x86 blade servers (and even more pleased to see some were ours). With their share rising from 5% to 75% in about five years, the intrusion of x86 servers into the Top 500 has been rapid and is replete with parallels to the eventual fortunes of dinosaurs at the hands of climate change and competition with more adaptable mammals. The top of the list is fascinating, a geekfest extrodinaire. But what really caught my eye was the composition of the bottom 100 of the list.

The top of the list is populated by machines that nobody in the real world could reasonably expect to be owned by real companies. In the top 50 there appear to be 4 machines that are not owned by academic or governmental research organizations or three-letter agencies, and of these four, three are owned by the vendors who make them. When we look at the last 100, the pattern changes, and we get a real sense of the impact that low-cost compute cycles has on the world. Almost all of these systems are medium-size clusters, dominated by IBM and HP blades, and the majority of them are in private hands who are putting the power of HPC clusters to work - from designing better beer cans to inventing cutting-edge pacemakers. Considering that #500 on today’s list offers substantially more performance than the top of the list 10 years ago, the degree to which real companies have found HPC to be a profitable and cost-justified investment is startling.

Will the trend continue? Hard to say with absolute certainty. An asteroid might strike the earth, or the sun might go nova. But failing that, it seems a certainty that real users will continue to absorb whatever capacity can be generated by the vendors as the price per cycle continues to decline. A recent survey of HPC users by IDC found that over 80% had problems that required 5 – 100 times the capacity they currently had in place. In the end we will all reap the benefits. Product design for basic consumer products, for example, has become incredibly sophisticated do to application of HPC – a seemingly mundane thing like making the most cost-effective beverage container may require detailed modeling with dozens of significant variables.

So expect better beer cans, pacemakers and airplanes as the HPC becomes a ubiquitous corporate resource and the last of the dinosaurs give may to the x86 mammals.