• United States

Fujitsu delivers exascale supercomputer that you can soon buy

News Analysis
May 27, 20203 mins
Data Center

Deployment will run through next year but Fugaku is already being put to work. Against Covid-19, naturally.

supercomputer / servers / data center / network
Credit: MaxiPhoto / Getty Images

Fujitsu has delivered all the components needed for a supercomputer in Japan that is expected to break the exaFLOP barrier when it comes online next year, and that delivery means that the same class of hardware will be available soon for enterprise customers.

The supercomputer, called Fugaku, is being assembled and brought online now at the RIKEN Center for  Computational Science. The installation of the 400-plus-rack machine started in December 2019, and full operation is scheduled for fiscal  2021, according according to a Fujitsu spokesman.

All told, Fugaku will have a total of 158,976 processors, each with 48 cores at 2.2 GHz. Already the partially deployed supercomputer’s performance is half an exaFLOP of 64-bit double precision floating point performance and looks to be the first to get to a full exaFLOP. Intel says  its supercomputer Aurora being built for the Department of Energy’s Argonne National Laboratory in Chicago will be delivered by 2021, and it will break the exaFLOP barrier, too.

An exaFLOP is one quintillion (1018) floating-point operations per second, or 1,000 petaFLOPS.

Fujitsu announced last November a partnership with Cray, an HPE company, to sell Cray-branded supercomputers with the custom processor used in Fugaku. Cray already has deployed four systems for early evaluation located at Stony Brook University, Oak Ridge National Laboratory, Los Alamos National Laboratory, and the University of Bristol in Britain.

According to Cray, systems have been shipped to customers interested in early evaluation, and it is planning to officially launch the A64fx system featuring the Cray Programming Environment later this summer.

Fugaku is remarkable in that it contains no GPUs but instead uses a custom-built Arm processor designed entirely for high-performance computing. The motherboard has no memory slots; the memory is on the CPU die. If you look at the Top500 list now and proposed exaFLOP computers planned by the Department of Energy, they all use power-hungry GPUs.

As a result, Fugaku prototype topped the Green500 ranking last fall as the most energy efficient supercomputer in the world. Nvidia’s new Ampere A100 GPU may best the A64fx in performance but with its 400-watt power draw it will use a lot more power.

Working to fight COVID-19

While construction marches on, RIKEN CCS and the Japanese Ministry of Education, Culture, Sports, Science and Technology have already started using the functioning parts of Fugaku to perform computations needed in research to fight the coronavirus. The projects it is working on include research into the characteristics of the virus, identifying potential drug compounds to combat the virus, research into diagnosis and treatment, insights into the spread of infections and its socio-economic impact.

It’s not the only supercomputer being turned against COVID-19. Government and industry organizations with supercomputers have  joined together in the effort, and CERN, the European nuclear research organization, redeployed a soon-to-be retired supercomputer with more than 100,000 cores on Folding@Home, the distributed-computing project that’s seeking a way to thwart the virus’s entry into human cells.

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