Fugaku was jointly developed by Fujitsu and the RIKEN Center for Computational Science in Kobe. Fugaku is the successor to the K computer at RIKEN, the first machine to break the ten petaflop barrier in 2011. Fugaku became the fastest supercomputer in the world according to the Top500 published June 2020.
Once deployed, Fugaku will be comprised of over 150,000 Fujitsu A64FX Arm processors connected via Fujitsu’s Tofu Interconnect D – a 6D mesh torus architecture named after Torus Fusion (Tofu for short).
The A64FX is impressive by any measure. It features a state-of-the-art 7nm process and implements the Armv8.2-A instruction set. The A64FX features 48 cores in 4 NUMA domains. It is also the first processor to implement the SVE (Scalable Vector Extension) instruction set. While 100% Arm compatible, the A64FX reflects Fujitsu’s heritage in SPARC64, adding superscalar capabilities, out-of-order execution, and branch prediction capabilities. Cray (now part of HPE) will also use Fujitsu A64FX processors in its Apollo 80 for customers in the US, Europe, and other countries.
In November 2019, Fujitsu announced the PRIMEHPC FX700, a modular air-cooled system based on the same technology as Fugaku, and can be easily deployed in commercial HPC environments.
In March of 2020, Univa completed testing on an FX700-based cluster verifying functionality with Univa Grid Engine 8.6.8 for Arm on Red Hat Enterprise Linux (RHEL) 8.0.
According to Akihiko Fujino, Fujitsu Limited, “Univa’s embrace of the PRIMEHPC FX700 platform is an important development for the Japanese HPC community. We have a long history working with Univa Grid Engine at supercomputing centers, including National Institute of Advanced Industrial Science and Technology’s ABCI, and look forward to continued collaboration.”
While Univa has long supported Arm-based systems, Univa strengthened its commitment in November, announcing support for NVIDIA GPU scheduling on Arm systems. This coincided with NVIDIA’s announcement of its Arm Preview Toolkit at SC19 in Denver.
“We see the PRIMEHPC FX700 as an important platform for Univa,” says Fritz Ferstl, Univa’s CTO. “These new Fujitsu systems, along with Cray’s decision to OEM Fujitsu A64FX processors, speak to both the strength of the technology and the growing importance of Arm platforms in HPC around the world.”
Univa Grid Engine is a compelling solution for heterogeneous HPC data centers. It provides excellent performance and scalability on-premise and in multiple clouds. Customers can deploy Univa Grid Engine across multiple architectures, including Arm, Intel, AMD, Power, and various GPUs, while supporting multiple operating environments, including Microsoft Windows.
To learn more about Univa Grid Engine visit http://www.univa.com