The Fugaku virtual supercomputer can now be run on the AWS cloud

The Japanese scientific group RIKEN Center for Computational Science presented a virtual version of its Arm supercomputer, which can be deployed in the AWS cloud. According to The Register, the supercomputer was considered the most powerful in the world in 2020 until it was surpassed by Frontier’s first exascale machine two years later.

Image source: RIKEN

The center intends to make it easier for those who wish to use the Fugaku system, which is why RIKEN decided to create a virtual twin capable of running in the cloud or even on supercomputers owned by other companies. Representatives of the center reported that building a machine from 160 thousand nodes is not enough, because software solutions are also needed. In other words, the Fugaku HPC software ecosystem has been completely reproduced in the cloud, which includes a lot of Arm-optimized packages and specialized software.

The first version of Virtual Fugaku is available as a Singularity image. It is designed to run on Amazon Graviton3E Arm processors, which are optimized for HPC/AI tasks. Like the Fujitsu A64FX processors used in Fugaku, they offer Scalable Vector Extension (SVE) instructions. The main OS is RHEL 8.10. The software is built using GCC 14.1 and the OpenMPI library, which supports EFA. Amazon is extremely pleased with the choice of AWS as the base platform for Virtual Fugaku.

Image source: RIKEN

In the future, it is possible that Virtual Fugaku will be ported to other architectures, but no matter what platforms it is transferred to, RIKEN hopes that the instances will “continue the work” of their parent. The researchers said the results of using Fugaku, including developments related to disease control, new materials and drugs, are well known. During operation, specialists gained a wealth of experience in handling the supercomputer and intend to share it with society.

RIKEN even considers Virtual Fugaku as a standard platform for using HPC software solutions – if supercomputing centers around the world adopt this format, users will appreciate the richness of the software library. However, some experts believe that such a concept is not entirely viable – HPC tasks often involve the use of equipment optimized for specific purposes, so it is unlikely that one software platform will suit all stakeholders.

admin

Share
Published by
admin

Recent Posts

FTC’s ‘One-Click Unsubscribe’ Rule Delayed Again, But Not for Long

The US Federal Trade Commission (FTC) has delayed a rule that would require companies to…

10 hours ago

KIBORG: left – crown, right – augmented. Review

Played on PC Developers from Sobaka Studio have built a reputation for themselves as authors…

12 hours ago

A simple and reliable speedometer for satellites has been invented in the USA

Satellites move in orbit at speeds of thousands of kilometers per hour, and without precise…

16 hours ago

Schoolchildren are starting to stick metal objects into Chromebook ports en masse for TikTok likes

A dangerous new TikTok challenge has gone viral in which American schoolchildren are deliberately damaging…

16 hours ago

Apple is developing powerful Baltra processors for AI servers, as well as consumer M5, M6 and M7

Apple is developing processors for data centers that will serve requests from Apple Intelligence artificial…

17 hours ago

Despelote — goo-o-o-o-o-o-o-o-o-o-ol! Review

One of my first memories (or perhaps the very first one – is it possible…

2 days ago