Who’s got the fastest chip — IBM? Intel ? Sun
? AMD
?
Those of you who answered ClearSpeed Technology can give yourselves a pat on
the back. That’s the claim anyway. The fabless semiconductor company plans
to demonstrate its CXS600 dual-chip board, running at 50 Gigaflops and a mere
25 watts, at the High Performance on Wall Street conference Monday
in New York.
A single Gigaflop equals a billion floating-point computer instructions per second, and it’s typically used to measure supercomputer performance.
Designed for computing or math-intensive applications, the board fits
standard PCI-X slots for workstations or servers or a cluster. Much as a
graphics coprocessor speeds the performance of video games, ClearSpeed is
designed to do the same in such areas as financial, life sciences,
computational chemistry, biology and computer-aided design for fluid dynamics.
“There is no one in the ball park close to our performance,” Mike Calise,
president of ClearSpeed, told internetnews.com. “We’ve been endorsed
by both Intel and AMD as a means to solve fundamental problems with heat and
power dissipation.”
ClearSpeed has not yet announced who will be selling boards based on the
CXS600, but Calise said there would be announcements from “major OEMs” this
year and next. Pricing for add-in boards will be up to the individual OEMs as their products come to market. “I can guarantee you the
prices per gigaflop will be unmatched in the industry,” said Calise, adding that a good estimate of pricing would start in the low five figures.
Some high-performance specialty chips have required developers to write
new software, an expensive, time-consuming process ClearSpeed means to
avoid.
“ClearSpeed provides a library interface that enables any of the
standard x86 compute-intensive packages to run without ISVs
making any changes,” said Nathan Brookwood, analyst with Insight64. “You combine that with the performance — that’s pretty
cool stuff.”
The CXS600
supports standard calls to the low-level math functions used by applications such as
MatLab by The Mathworks and Mathematica by Wolfram Research.
“If you look at all the multicore announcements coming out, the industry
is inherently going to parallel processing, because you can get phenomenal
results,” said Calise.
Way beyond dual-core, the ClearSpeed chip features 96
processors running at “only 250 MHz” each, but gain super speed performance
running in parallel.
The need for faster mathematical processing is particularly acute in
corporations having to deal with the government’s complex Sarbanes-Oxley
regulations. Calise also mentioned research into new drugs and the human
genome as other likely applications.
“And then you have the Wall Street
traders who’ve seen a doubling of their data sets. They’re looking at grid
workstation level.”
IBM has said its forthcoming cell processor will be theoretically capable of running as fast as 256 gigaflops per
second. It’s slated to first appear in Sony’s PlayStation 3 game system
later next year, although it’s also expected to have commercial applications.
IBM is the foundry supplier to ClearSpeed.
“Cell is designed as a 32-bit
device and as a standalone processor, so it’ll be competing more against
RISC solutions,” said Calise. “We’re 64-bit and work integrally with the
main CPU, so we’re really in a different market.”
