Nvidia nets key supercomputer trophy for Tesla business

The Titan supercomputer, housed at the Oak Ridge National Laboratory, will ultimately feature 18,000 Nvidia Tesla GPUs.
Written by Larry Dignan, Contributor

Nvidia's Tesla chip will power the world's fastest supercomputer in what could amount to a nice trophy win for GPU-powered servers.

On Tuesday the company announced that the U.S. Department of Energy will deploy a Cray XK6 supercomputer dubbed Titan that will be able to deliver 20 petaflops at peak performance. In a nutshell, Titan will be twice as fast as the K computer in Japan. The K computer is the fastest system to date.

The Titan supercomputer, housed at the Oak Ridge National Laboratory, will ultimately feature 18,000 Tesla GPUs. Oak Ridge is currently upgrading its existing Jaguar supercomputer with 960 Tesla GPUs, which will be companion processors for Titan. In 2012, Oak Ridge will start the larger Titan deployment.

So what does Nvidia's win mean? Beyond a little chest thumping not much. Nvidia's Tesla business remains a work in progress and the real payoff will be years away.

Nvidia CEO Jen-Hsun Huang outlined the plans for Tesla last month at a Citi investment conference. He said:

Our third GPU brand is Tesla. It's a revolutionary processor, whereby we generalize the parallel computing capability of 3D graphics to make it possible for you to accelerate parallel computing applications, molecular dynamic simulations, computational biology, computational finance as used here in Wall Street to do risk analysis much, much more quickly. We've shipped today about 150,000 processors. That represents about 20 times the computational capability of the fastest supercomputer we helped install in China.

If you were to take 150,000 processors, the computational capability we've shipped in just about a year or so represents the aggregate computational capability of all the Top 500 supercomputers in the world, just to put that in perspective, revolutionary speed up of computation.

My expectation and our vision is that in some 10 years’ time, these processors that are used in computing clusters for high-performance computing, which represents about a $10 billion market and/or another way to think about it, about 4 million nodes per year are sold into technical computing. All of those processors would be massively parallel someday. And so that represents -- we've shipped 150,000 units of this market and someday, the annual market is a few million units. So, big growth opportunity for us there.


Editorial standards