This site may earn affiliate commissions from the links on this page. Terms of use.

TheMachineHP

Hewlett-Packard Enterprise is prepping the most powerful ARM-based supercomputer ever built for installation at Sandia National Laboratories later this twelvemonth. The organization, codenamed Astra, will exist built around HPE's Apollo 70, with 2,592 dual-socket nodes and a total of 145,152 cores.

But unlike about supercomputers today, including virtually the entire TOP500 listing, Astra isn't congenital effectually the x86 compages. Information technology shares that stardom with only 24 other systems on the existing TOP500, with the other 476 systems based on Intel or AMD hardware. Astra will pack 5,184 ARM CPUs — each a 28-core Thunder X2 CPU built by Cavium. Each CPU is clocked at 2GHz, for a full of 2.3 height petaflops in theory, a rank which would put the new organisation roughly around #87 on the latest TOP500 list released in Nov 2022. By the time Astra is fully online new lists volition have been published, which is probably why the press materials mention the more than generic "Superlative 100" of the TOP500 rather than attempting to land where, exactly, the new system will rank.

Astra volition be "hands down exist the earth's largest ARM-based supercomputer always congenital," Mike Vildibill, VP of Advanced Technologies Grouping at HPE, told ZDNet. "The regime views ARM as 1 of several microprocessors that are important for achieving exascale in the futurity."

astra_rendering-800x316

Astra rendering, courtesy of HP

Keeping workloads local and moving data minimally are key goals in the push to build exascale-class computing hardware, given the tremendous energy toll of moving information from Point A to B. "We tin run into conspicuously that the corporeality of power required to motion data within the system is an gild of magnitude greater than the amount of ability needed to compute that information," explained Vildibill.

In recent years, the US government has doled out $258M towards exascale calculating in grants to a number of companies, including AMD, Intel, HPE, Cray, IBM, and Nvidia. Overall ability consumption volition be 1.2MW, an amount the TOP500 printing release characterizes as "respectable energy efficiency." The system will be backed up past 350TB of storage and is intended to search for new methods of managing America's crumbling nuclear armory. The installation is as well a kind of test run for ARM hardware in these kinds of situations overall, and to examine its performance characteristics in scenarios that ARM hasn't historically been used to examination.

Sandia's calculations are reportedly bandwidth-limited, and some of Cavium'southward Thunder X2 CPUs take up to eight DDR4 memory controllers, capable of providing up to 170GB/s of bandwidth per socket. HP is making this push equally part of its overall Retention Driven Calculating compages approach, in which the company is emphasizing the amount of memory channels and total connected RAM it can offering with its servers. In this case, you don't technically need to get with ARM to maximize per-socket memory bandwidth — AMD also uses upward to eight channels in its Epyc line of CPUs — but the movement to deploy a test ARM arrangement is also important for validating that ARM hardware and servers are up to the challenge of serving in this environment.