Jaguar knows that snob appeal sells cars.
And apparently, it also sells supercomputers. Just ask Cray: With the recent release of its new X1 supercomputing system, available with up to 52.4 teraflops of peak computing power and with a starting list price of $2.5 million, Cray isn’t planning to waste its time on the bioinformatics masses. “Cray is going to be targeting the people who are doing the most serious new science in this field,” said Steve Conway, Cray’s vice president of corporate communications. “If you’re primarily running Fasta and Blast and things like that today, you’re clearly not doing serious new science, because those are reductionist codes,” he added.
In other words, you can keep your Blast farm.
Lori Kaiser, vice president of marketing and strategic planning, used a fitting analogy to describe how Cray’s technology compares to cluster-based approaches: “If you want to go faster in a car, would you buy two cars or would you buy a more tightly designed car to go faster? It’s not about theoretical peak performance, it’s about what you can get done and how fast you can get it done.”
Cray expects to find strong demand for the X1 in the life science market for modeling and simulation applications that can take advantage of its memory bandwidth — the speed at which it can exchange data between processors. According to the company, the main memory bandwidth per CPU of the X1 is five times faster than its closest competitor, IBM’s P690. Applications in cellular modeling, protein-protein interactions, modeling systemic reactions to a drug, computational drug design, and modeling epidemics and pandemics would all benefit from the high-bandwidth communication in the Cray architecture, Kaiser said.
The Army is already using an early production X1 system to simulate “dropping an incendiary device in the city of Atlanta, and how that spreads in density across a cityscape given a certain climate, in order to determine how to evacuate and how to treat certain block areas,” Kaiser said.
An early production version of the X1 is also in place at Germany’s Konrad Zuse Center for Information Technology and Bielefeld University, where researchers are using the system to develop new algorithms for EST clustering, sequence alignment, protein structure prediction, and other compute-intensive tasks.
The Institute for Systems Biology and BioNumerik Pharmaceuticals — already users of the Cray SV1 — are also in talks with the company about purchasing the X1, Conway said.
Only five production systems were installed prior to the X1’s launch in mid-November, and Cray deliberately tried to keep a low profile in order to avoid overhyping the upcoming system, Kaiser said.
Despite a significant lag in new product launches and the recent rise of low-cost clustered computing approaches, Cray received an unexpected burst of interest in its technology this April when Japan’s Earth Simulator, built by NEC, came online. The Earth Simulator — as fast as the top 20 US supercomputers combined — validated Cray’s architecture, which is similar to NEC’s.
Cray said it is the only US computer manufacturer to use comparable technology to NEC, and Oak Ridge National Laboratory is currently evaluating the X1 under a DOE project to build a supercomputer to match the performance of the Earth Simulator.
In spite of the welcome surge of interest in Cray’s technology, Conway said that the company still has a tough time convincing some customers that its high-end system offers actual benefits over other approaches, especially in the life science market. “In general, the pharmaceutical industry doesn’t have a good understanding of computing,” he said. “The scientists know the science, but a lot of them think, ‘Well, a computer’s a computer.’ Like light or plumbing.”
However, once the X1 gets in the hands of a researcher who can really benefit from the system’s architecture, “a light goes off,” he said. Cray is counting on those experiences — much like stepping out of a Hyundai to test drive a Jaguar — to sell its sporty new system into the very top end of the life science research market.