Hewlett Packard Enterprise Signs Big Supercomputer Deal with Defense Department
The Defense Department is paying $57 million to Hewlett Packard Enterprise for supercomputers that it plans to use for tasks like designing helicopters and weather forecasting.
The Air Force Research Laboratory and Defense Department’s Supercomputing Resource Center at the Wright-Patterson Air Force Base near Dayton, OH will receive four HPE SGI 8600 supercomputers, as part of the deal. Another three of the same supercomputers will be installed at the Navy Department of Defense Supercomputing Resource Center in southern Mississippi.
The Defense Department bought the supercomputers as part of its so-called high-performance computing modernization program, said the program’s chief of staff, Kevin Newmeyer. The program, created in 1992, is intended to ensure that the Defense Department consistently has the most powerful computers for tasks like designing weapons, aircraft, and analyzing weather patterns so that Navy ships can navigate more safely, he said.
The deal, which includes consulting and maintenance, is noteworthy for HPE (HPE) because the company recently bought supercomputer company SGI in 2016 for $275 million, and inherited its existing contracts.
Because the Defense Department is required to buy supercomputers made in the U.S., its choices are limited to only a few companies. Currently, the agency has some Cray supercomputers, SGI supercomputers, and older IBM (IBM) machines, Newmeyer said.
Some of the Defense Department’s plans for the new supercomputers involve developing new helicopters, Newmeyer said. Although contractors like Boeing (BA) and Lockheed Martin’s Sikorsky typically develop and build aircraft for the military, the department also contributes to the aircraft design plans to save time, he explained.
The supercomputers will help the Defense Department simulate wind tunnels for testing the software models of helicopters prior to them being built, Newmeyer said. This limits the chance of errors once the physical helicopters are made and eventually tested in actual wind tunnels.
“We create digital representations and images that matches all the curves and everything else,” Newmeyer said. “Then we take the digital image and fly it using the supercomputers.”
The Defense Department needs supercomputers to crunch the “literally billions of data points” related to projects like designing helicopters, Newmeyer said. Despite the power of supercomputers, these kinds of heavy-duty tasks can still take days.
The Defense Department also wants to incorporate artificial intelligence-related tasks like machine learning to help speed the design process. And like many of today’s companies, the military is turning to so-called graphics processing units to help it do so, Newmeyer said. Businesses like Nvidia have benefited in recent years as companies use those GPUs to help with A.I.-related tasks like deep learning.
HPE supercomputing and A.I. head Bill Mannel said the latest version of the company’s supercomputers include more Nvidia GPUs than they ever have and HPE is exploring different types of specialized chips like so-called field programmable gate array (FPGA) chips to help improve the overall performance for certain tasks.
“At one time, you could always count on the CPU being able to give a boost of performance each year,” Mannel said. That’s become more of a challenge.”
Get Data Sheet, Fortune’s technology newsletter.
Although the Defense Department is buying this batch of supercomputers, it could eventually shift some of its computing to cloud computing, in which case it would rent computing capacity from companies like Amazon and Microsoft. In terms of cloud computing, the military has security concerns and wants to keep sensitive data “in house to a large extent” rather than crunching it on a third party’s computers, Newmeyer said.
That said, if those security and regulatory concerns are addressed, the Defense Department could potentially use the power of thousands of computers tethered together by an outside contractor instead of one beefy machine.
“In the not too distant future,” Newmeyer predicts, “the cloud will be able to do supercomputing jobs at a competitive price.”