SAN JOSE, Calif. -- Nvidia Corp. announced its next major leap in technology, vowing to deliver chips that speed up an array of computing chores as well as advancing its traditional stronghold in graphics.
The Silicon Valley company demonstrated a new chip design, dubbed Fermi, that it said will target scientific and business applications beyond conventional chores such as generating images in videogames. At three billion transistors, Nvidia said its new chip would be the most complex ever designed.
But the company -- known for chips known as GPUs, for graphics processing units -- is discussing its latest technology later than its chief rival, Advanced Micro Devices Inc. AMD this month began selling a 2.15-billion transistor GPU that has been getting positive reviews.
Jen-Hsun Huang, Nvidia's chief executive, was not specific about the arrival of chips based on the Fermi design. But he noted the company has working chips now, and usually takes a few months after that to begin shipping products to the market.
"Nobody likes it when the competition has a product and we don't," Mr. Huang said, in a press conference following his speech at a technical conference Nvidia is hosting here. But he stressed that the company's strategy in addressing markets beyond graphics is different than AMD's, and predicted the delay won't have much impact.
"This is likely to be the most successful GPU we have ever announced," Mr. Huang said.
Nvidia has for several years been pushing the use of GPUs in new markets. But Nvidia's products, though they have many calculating engines that can work together in parallel, still lacked features demanded for some business and scientific applications.
Mr. Huang stressed that Fermi is based on a fundamentally new design that addresses such shortcomings. He said the new chip will "treat computer graphics and parallel computing as equal citizens."
Among other things, the technology offers an eight-fold speedup in so-called double precision arithmetic, which is needed for applications such as linear algebra, numerical simulation and quantum chemistry, the company says.
Fermi also has features to make it much more easy to program, using the programming language C++ and development software such as Microsoft Corp.'s Visual Studio, Huang said. GPUs are generally considered much more difficult to program than microprocessors, which handle most of the calculating chores in computers today.
Mr. Huang demonstrated said the new technology also will have a big impact in videogames and related markets. For one thing, the company plans to exploit a technology called ray-tracing that has mainly been used so far in movie-making rather than interactive games. "This year we are going to bring real-time ray-tracing to the world," he said.
In one endorsement of the company's new strategy, Oak Ridge National Laboratory said it would use a future chip derived from the Fermi design in a large supercomputer that would be ten times faster than today's fastest scientific systems. The lab has so far mainly built supercomputers using microprocessors from Intel and AMD.
It's interesting they 'launch' this architecture now. When I was there, the theory was you never pre-announce -- unless you're in second place as you don't have anything to lose in that case.
Some look at this as a new chip announcement, like we're all going to be playing Crysis on GT300s in Q1. Maybe that's true, but I don't think that's the motivation here.
Jensen is painting a bullseye squarely on Intel. He's defined a model where he co-exists -- he doesn't take away their existing business, he just wants to limit (and take away any of) their HPC upside. With the Larrabee mis-steps, they are clearly looking at a window of opportunity to dethrown the 'compute' performance leader. This is a technology announcement with a loud and clear message: Intel, we're coming at you.
Yeah, I thought nvidia doesn't pre-announce stuff either. But guess Fermi is late, and this conference was about Tesla & Cuda....so they probably felt it was the right thing to do to inspire all those future apps.