Nvidia: Tesla K10 GPUs Strong In Defense, Other Verticals

At the International Supercomputing Conference taking place this week in Hamburg, Germany, Nvidia said its Tesla GPUs can be found in nearly 60 of today's Top 500 supercomputing platforms, compared to just 10 in 2010. Nvidia's K10 GPUs, which are based on its next-gen Kepler architecture and were unveiled last month, are driving much of this adoption, the chip maker said, as they are three times more energy efficient and pack twice the processing punch of prior-generation GPUs based on its Fermi architecture.

Nvidia’s Kepler architecture also allows for two Tesla K10 GPUs to be used on a single compute accelerator board, enabling them to reach aggregate speeds of up to 4.58 teraflops and memory bandwidth of up to 320 GBps. This is possible, said Roy Kim, senior product manager of the Tesla HPC business unit at Nvidia, because of Kepler’s low-power model.

[Related: Nvidia Unveils New GTX 690 Kepler-Based GPUs ]

"The Tesla K10 actually does have two GPUs on board, and that’s actually what really defines Kepler for us -- power efficiency," Kim told CRN. "Because it’s so power-efficient we are able to put two on the same board."

The speed and memory bandwidth delivered with the new Tesla K10 GPUs make them especially valuable to the defense industry, where they are used in workstations to accelerate memory-intensive applications such as those for video analytics and stabilization. They are also being used in the oil and gas, media and entertainment and life sciences market, Kim said, to accelerate simulation and image rendering software.

In life sciences, for example, Kim said Tesla K10 GPUs are being used in place of massive CPU clusters to accelerate biomolecular simulation software used for medical research. According to Kim, a system running two Tesla K10 GPUs can deliver up to 66 nanoseconds of simulation compared to just 52 nanoseconds delivered by eight prior-generation Tesla M2050 GPUs in 2010.

While a difference of nanoseconds may seem trivial, it's significant for industries such as life sciences and defense that rely heavily on the accuracy on simulation software.

"In biomolecular science, adding a few more nanoseconds of simulation time can make a world of difference in the ability of researchers to study and better understand the behavior of complex biological systems," said Ross Walker, assistant research professor, San Diego Supercomputing Center, in a statement.

Kim told CRN that Nvidia channel partners stand to benefit from the rise in demand for Kepler-based GPUs in large vertical markets like medical research and defense, which tend to be segmented and hard to reach through direct sales alone.

"The defense segment is actually quite fragmented, so there are a lot of channel partners and VARs supplying to the defense industry," Kim said.

What's more, he continued, the defense segment is grappling with big data issues because of its reliance on data-intensive applications for video analytics. Nvidia partners, as a result, have an opportunity to help defense organizations handle big data with high-bandwidth K10 GPUs.

In addition to the new Tesla K10 GPUs, Nvidia will be unveiling its Tesla K20 GPU, optimized for high-performance computing in government and financial institutions, in the fourth quarter.