Americas

  • United States

Cray to license Fujitsu Arm processor for supercomputers

News Analysis
Nov 14, 20192 mins
Data Center

HPE's Cray will co-develop Fujitsu's A64FX CPU to meet the requirements of likely customers such as universities and national research laboratories.

K Computer supercomputer
Credit: Riken Advanced Institute for Computational Science

Cray says it will be the first supercomputer vendor to license Fujitsu’s A64FX Arm-based processor with high-bandwidth memory (HBM) for exascale computing.

Under the agreement, Cray – now a part of HPE – is developing the first-ever commercial supercomputer powered by the A64FX processor, with initial customers being the usual suspects in HPC: Los Alamos National Laboratory, Oak Ridge National Laboratory, RIKEN, Stony Brook University, and University of Bristol.

As part of this new partnership, Cray and Fujitsu will explore engineering collaboration, co-development, and joint go-to-market to meet customer demand in the supercomputing space. Cray will also bring its Cray Programming Environment (CPE) for Arm processors over to the A64FX to optimize applications and take full advantage of SVE and HBM2.

The A64FX was announced last year as the processor for Fujitsu’s next supercomputer, known as Post-K. The K supercomputer is a massive system at Japan’s RIKEN Center for Computational Science, based on the Sparc architecture. Fujitsu had a Sparc license from Sun Microsystems and made its own chips for the Japanese market.

A64FX is the first CPU to adopt the Scalable Vector Extension (SVE), an extension of Armv8-A instruction set architecture for supercomputers. SVE is focused on parallel processing to run applications faster.

The A64FX also uses HBM2, which has much greater memory performance than DDR4, the memory standard in servers. The A64FX has a maximum theoretical memory bandwidth greater than 1 terabyte per second (TB/s).

Fujitsu claims the A64FX will offer a peak double precision (64-bit) floating-point operations performance of over 2.7 teraflops. That pales in comparison to the 100 TFlops for an Nvidia Tesla V100, but the A64FX has a power draw of 160 watts vs. 300 watts for the Tesla.

However, there is more going on. The 32GB of on-chip HBM2 and high-speed interconnects mean a much faster internal chip, and in early tests, Fujitsu is claiming a 2.5-times performance improvement over the Sparc XIIfx chips used in the K computer.

The Cray supercomputer powered by Fujitsu A64FX will be available through Cray to customers in mid-2020.

Andy Patrizio is a freelance journalist based in southern California who has covered the computer industry for 20 years and has built every x86 PC he’s ever owned, laptops not included.

The opinions expressed in this blog are those of the author and do not necessarily represent those of ITworld, Network World, its parent, subsidiary or affiliated companies.

More from this author