dcsimg

IBM Adds Nvidia GPUs to High-Performance Servers

By Andy Patrizio (Send Email)
Posted May 19, 2010


Nvidia and ATI, the graphics arm of AMD, have been promoting their graphics processors as viable alternatives for compute-intensive tasks for some time, but so far, mainstream server vendors have been slow to bite. Until now. GPU processors will be used as high-performance calculation engines in number-crunching tasks. Have GPUs finally arrived in HPC?

IBM (NYSE: IBM) announced it would include Tesla processors from Nvidia (NASDAQ: NVDA) in its System x iDataPlex scale-out servers. The new servers are aimed at organizations that need to run tasks with massive amounts of mathematical computation, and while GPUs -- short for graphics processing units -- have been generally used to render advanced 2D and 3D graphics, at their heart, they're essentially floating point processors with hundreds of cores interconnected to do math computation.

For the past few years, Nvidia and ATI have been beating the drum for using GPUs in high-end business and scientific processing, for things like financial services, medical image processing, atomic simulation, oil and gas exploration and medical simulations.

IBM was first to do a hybrid system with its Roadrunner supercomputer for Los Alamos National Laboratory. But the hardware is a combination of core AMD Opteron CPUs and IBM's Cell processor, the processor used in the Sony PlayStation 3, which IBM has ceased developing. The effort proved enough to put the computer on the top of the Top 500 list of supercomputers for a short while.

In China, a new petaflop supercomputer was built with Intel processors and ATI Radeon video cards late last year as well, but that was a one-off design by the Chinese government. Now, IBM's deal with Nvidia marks a new stage for the burgeoning space, and comes as the first major OEM win for Tesla, which up to now Nvidia has sold on its own.

"Nvidia provides an innovative solution for customers who push the envelope in high-performance computing," Dave Turek, vice president of deep computing at IBM, said in a statement. "GPU acceleration provides performance boosts for many applications in energy exploration, science and financial services. It is among the significant emerging supercomputer technologies to watch in the years ahead."

The iDataPlex dx360 M3 is a two-socket server that uses Intel's Xeon processors and two Nvidia's Tesla M2050 GPUs. These Teslas are from Nvidia's new Fermi generation of GPU design: massive chips with more than three billion transistors. Nvidia claims that the iDataPlex servers will run up to 10 times faster than Xeon-only systems.

IBM launched the iDataPlex in 2008 as an alternative to rack-mount servers for large, scale-out computing environments. The company recently upgraded them to add storage and memory capacity as well as support for Intel's latest Xeons, the 5600 line.

IBM chose Nvidia's GPUs because of its support for OpenCL, a framework for writing applications that us both CPUs and GPUs.

Jon Peddie, president of graphics market research firm Jon Peddie Research, called the IBM-Nvidia deal "great news. This is verification and validation of the Tesla product as being a viable server supercomputer accelerator. Having a name like IBM really puts a stamp of approval on it, and you'll see other companies following that," he told InternetNews.com. "I expect to see Dell, HP and Sun all doing that. Not necessarily with Tesla, but adding GPU computing."

It's also good for Nvidia's rival, ATI, because it, too, has a strong play. The difference is that ATI supports OpenCL while Nvidia supports OpenCL plus its own CUDA effort. "CUDA is much stronger than OpenCL right now. It has a richer C language and compiler and supports Fortran," while OpenCL is not as mature and needs some work, Peddie said.

The new iDataPlex servers are expected in July. Pricing will be announced then.

Andy Patrizio is a senior editor at InternetNews.com, the news service of Internet.com, the network for technology professionals.

Follow ServerWatch on Twitter

Page 1 of 1


Comment and Contribute

Your name/nickname

Your email

(Maximum characters: 1200). You have characters left.