Sandia Nationwide Laboratories has introduced will probably be the primary Division of Vitality labs within the US to deploy the Fujitsu A64FX, the one ARM-based processor designed from floor up for HPC tasks and supercomputers.
Fujitsu is thought primarily for its enterprise laptops, tablets and desktops, however is a behemoth in its personal proper in the case of processors, having been within the enterprise for effectively over half a century.
Launched in 2019, the CPU has 48 cores, a theoretical peak efficiency of three.38 TFLOPS, runs at 2.2GHz and has 32GB HBM2 reminiscence on the die itself.
What makes it ideally suited for the HPC market is that it gives far increased bandwidth efficiency between reminiscence and the CPU – as much as 1TBps. Shifting information to and from the CPU is the most important impediment by far to what researchers discuss with as exascale computing.
What makes the A64FX much more thrilling is that Fujitsu desires the know-how to trickle all the way down to hyperscalers and main cloud computing giants in order that the plenty can profit too.
Given it’s primarily based on ARM structure, it may well (and has) run Linux distributions out of the field and even Microsoft Home windows.
It’s thought of a common objective CPU, however surpasses even GPUs from Nvidia and AMD on the all-important metric of efficiency per watt. Certainly, a 768-CPU prototype sits on high of the Inexperienced500 record – the leaderboard for supercomputers that ship probably the most energy per watt.
The A64FX was designed expressly to energy the successor of Japan’s fundamental supercomputer, the Ok, which was decommissioned again in August 2019.
Its substitute – the Fugaku – is anticipated to be 100 instances quicker when it launches later this 12 months, will run on a Linux distribution known as McKernel and can attain a staggering 400 petaflops. The purpose is for it to be the primary supercomputer to hit one exaflop when totally deployed with half one million processors buzzing.