Computerworld - AUSTIN -- The definition of a personal supercomputer goes something like this: It's inexpensive, can sit on a desk, plugs into a wall socket and is at least within jumping distance of the Top500 supercomputing list. By that measure, Nvidia Corp.'s new computer is one of the first arrivals in this emerging product category.
Nvidia today unveiled a workstation it calls the Tesla Personal Supercomputer at the Supercomputing 08 show here. The Tesla sports 960 cores, delivers almost 4 teraflops of performance and costs less than $9,995. It achieves that speed and price by using four graphics processing units (GPU), each of which has 240 cores.
"This really is the supercomputer on your desk," Dell Inc. CEO Michael Dell told attendees at the conference, which drew more than 10,000 science and commercial high-performance computing users, along with a slew of vendors hoping that the high-performance computing (HPC) market will be a bright spot in otherwise difficult tech economy. The attendance was a new record for this annual conference.
The development of HPC systems that rely on GPU processors has advanced, in part, because of the billions of dollars being used to develop systems for gamers. The contribution gaming has made to HPC development wasn't lost on those at this conference.
"That's the beauty of it," said Ian Watson, a chemist who uses high-performance systems for a pharmaceutical company he didn't want named. "The gamers of the world are paying for the development of that stuff. If we can hitch ourselves to that train as it thunders past, that's very attractive."
Dell has produced a workstation in its Precision line that uses the Nvidia GPU chip, and a number of other vendors are producing systems based on Nvidia's parallel computing architecture.
This is type of system is designed to support applications that require a high degree of parallelization, such as visualizations, seismic studies, biomedical research and product design, among others. Nvidia released a software development kit about a year ago so independent software developers could build applications to run on the new system.
In September, Cray Inc. released a Windows HPC server machine, the CX1, which relies on Intel chips for compute capability. The CX1 cost about $25,000.
"Graphics is the ultimate parallel computing application," said Andy Keane, the general manager of Tesla Computing Products. Keane demonstrated the system on the trade show floor.
The computing ability the Tesla offers is typically available to most users through the use of a cluster or some shared-computing resource.
Just what constitutes a supercomputer remains a rapidly moving target, with the fastest defined by the Top500 list maintained by academics in the U.S. and Europe. The Nvidia personal supercomputer would have been ranked among the top 20 fastest systems just five years ago, but it isn't fast enough to make the most recent list. The 500th ranked system is rated at about 12.6 teraflops.
HPC has been one of the fastest, if not the fastest, growth areas in the tech sector. There was very little evidence of economic gloom here at the 20th annual Supercomputing conference. One person who did refer to the current economy was Michael Dell, who said he expects to see some corporate budget cuts even as he argued that the importance of HPC capabilities in an HPC server has never been greater.
Read more about Mainframes and Supercomputers in Computerworld's Mainframes and Supercomputers Topic Center.
Free Insider GuideCopyright © 1994 - 2012 Computerworld Inc. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of Computerworld Inc. is prohibited. Computerworld and Computerworld.com and the respective logos are trademarks of International Data Group Inc.
