X

Arm chips, like those in your phone, will power massive Astra supercomputer

Hewlett Packard Enterprise's Astra machine will study nuclear weapons' safety and reliability. It'll be fast, but not the fastest.

Stephen Shankland Former Principal Writer
Stephen Shankland worked at CNET from 1998 to 2024 and wrote about processors, digital photography, AI, quantum computing, computer science, materials science, supercomputers, drones, browsers, 3D printing, USB, and new computing technology in general. He has a soft spot in his heart for standards groups and I/O interfaces. His first big scoop was about radioactive cat poop.
Expertise Processors, semiconductors, web browsers, quantum computing, supercomputers, AI, 3D printing, drones, computer science, physics, programming, materials science, USB, UWB, Android, digital photography, science. Credentials
  • Shankland covered the tech industry for more than 25 years and was a science writer for five years before that. He has deep expertise in microprocessors, digital photography, computer hardware and software, internet standards, web technology, and more.
Stephen Shankland
2 min read
The Sandia Science & Technology Park, part of Sandia National Laboratories headquartered in Albuquerque, New Mexico.

The Sandia Science & Technology Park, part of Sandia National Laboratories headquartered in Albuquerque, New Mexico.

Sandia National Laboratories

For years, Intel has dominated the market for processors powering supercomputers. But a new system from Hewlett Packard Enterprise will be built this year with a rival design from processor maker Arm, better known for smartphone chips.

The system, called Astra and arriving at Sandia National Laboratories in Albuquerque, New Mexico, late this summer is built from 2,592 servers, each with two ThunderX2 Arm processors built by Cavium. Each processor has 28 cores, and each server is about 100 times faster than a modern phone, Sandia said.

It's the biggest Arm-based machine yet, capable of performing quadrillions of calculations per second, HPE said Monday. That's quite an achievement for a chip family that got its start with handheld computers like the Apple Newton and rose to prominence powering just about every phone today, from Apple's iPhones to competing models using Google's Android software.

Why mess with Arm chips when Intel Xeon processors are the top pick for supercomputing today? Arm-based servers are more economical with power and can be packed more densely, and HPE likes their memory performance, too, the company said Monday.

An illustration of the Astra supercomputer, powered by Cavium-built Arm processors and built by Hewlett-Packard Enterprise at Sandia National Laboratories headquartered in Albuquerque, New Mexico.

An illustration of the Astra supercomputer, powered by Cavium-built Arm processors and built by Hewlett Packard Enterprise at Sandia National Laboratories.

Sandia National Laboratories

Astra is the first of a "potential series of advanced architecture prototype platforms" in Sandia's Vanguard program to explore new computing designs for the US nuclear weapons program, Sandia said.

Even Astra will be far from the fastest machine around. But it's still a significant step in the push by the US and Chinese governments to reach "exascale" supercomputing -- an exaflop, or quintillion calculations per second.

The fastest machine on the Top500 supercomputer list, a record updated twice yearly by a group of researchers in the field, is 93 petaflops, about a tenth of the way toward the exaflop milestone. It's likely IBM's new Summit supercomputer at Oak Ridge National Laboratory will at least double that when the new Top500 list is released later in June. It uses IBM's Power9 processors, but 94 percent of the Top500 machines use Intel processors.

Astra's theoretical peak speed should be 2.3 petaflops. You'd need the entire population of Earth to perform 300,000 calculations each second to match that. However, the Top500 supercomputers are ranked on sustained performance, a lower number than the theoretical peak.

Part of the purpose of Astra will be to see how well this theoretical performance turns into actual research performance, said Mark Anderson, program director for the Advanced Simulation and Computing program, funded by the Energy Department's National Nuclear Security Administration.

First published 11:56 a.m. June 18.

Correction, 1:42 p.m. June 22: Corrects the number of processor cores for each Cavium chip. Each processor is a 28-core model.

CNET Magazine: Check out a sample of the stories in CNET's newsstand edition.

'Hello, humans': Google's Duplex could make Assistant the most lifelike AI yet.