First impression on unpacking the Q702 test unit was the solid feel and clean, minimalist styling.
NEC hopes new supercomputer can challenge IBM, Cray
- — 29 October, 2007 08:19
With the launching of its a new vector supercomputer, NEC executives said they are racing after industry leaders IBM and Cray to recapture a piece of their faded supercomputer glory.
Two industry analysts, however, said it will take more than a vector supercomputer -- no matter how well designed and powerful -- to do that.
"NEC says they plan to go after IBM, but I think they need to focus on clusters," said Charles King, an analyst at Pund-IT. "Over the last couple of years, we've seen a renaissance in supercomputing. Clustered systems have been the prime motivator for that, putting supercomputers into the hands of companies that couldn't afford them a decade ago. The supercomputing market has moved from vectors to clusters."
NEC, though, does see a future in the vector market, judging by today's unveiling of the SX-9 machine that NEC promises will hit peak processing performance of 839 TFLOPS.
Dennis Lam, principal engineer at NEC, said that the SX-9 processing speed has yet to be benchmarked, but he nonetheless expects it to become the world's fastest computer.
King, however, said: "The NEC announcement is a bit overblown in that the company claims to have created the world's fastest vector-based system. That's probably true, but the 839 TFLOPS performance that NEC claims for a fully deployed SX-9 system significantly lags IBM BlueGene's sustained PFLOPS performance."
The BlueGene/L system, which is installed at the Lawrence Livermore National Laboratory, is currently ranked as the world's fastest supercomputer.
NEC also noted that the SX-9 features what it calls the world's first CPU capable of a peak vector performance of 102.4 GFLOPS. The company claims the new supercomputer closes in on the PFLOPS range, or 1,000 trillion calculations per second.
Steve Conway, an analyst at IDC, noted that the SX-9 is "very interesting" and well designed, but agreed that demand for vector systems is declining.
"The NEC machine is a vector supercomputer and that can't be beat for certain types of problems," said Conway. "They're very useful in weather predictions and in designing automobiles. But the number of applications for vector supercomputers has really shrunk. It's more specialized than general run-of-the-mill supercomputers these days. More than a decade ago, vector supercomputers were the dominant species of supercomputing. That's just no longer the case."
He added that 10 years ago, many applications for tasks like weather predictions and auto design were built for vector supercomputers. Demand for such vector-based software has since fallen considerably because of the growth of cluster systems based on standard microprocessors, Conway said.
King pointed out that when Top500.org's Top 500 Supercomputer list was first launched in 1993, 66.8% of systems were vector based. By 2000, however, that number dropped to 12%, and then to 3.6% in 2005. On the most recent Top 500 List, King noted, vector-based systems made up 1.2% of the systems represented.
The only NEC system currently on the list, the 20th ranked SX-6 Earth Simulator held the top position for a couple of years until it was knocked out by the BlueGene/L system, King reported.
NEC's Lam agreed that the industry is trending toward clusters, but he maintains that the vector is still a "useful platform" for many market segments.
Conway noted that NEC's SX series of supercomputers have been popular in Japan and in Europe. In the U.S., however, companies buying vector machines generally turn to Cray, he said.
"If your applications are written to take advantage of vector machines, then the arrival of a new generation of a NEC SX machine is very exciting," said Conway.
"NEC has really, from generation to generation, pushed the technology forward in a serious way. The NEC design is very impressive by any standard. This is a very advanced piece of technology. Its appearance in the marketplace is going to be very valuable to a subset of customers around the world, but it's not as general purpose as a Linux cluster machine. The market for it is probably in the single digits," he added.