Sign up for our newsletter and get the latest HPC news and analysis.
Send me information from insideHPC:


ARM to Power New RIKEN Supercomputer

armARM processors will provide the computational muscle behind one of the most powerful supercomputers in the world, replacing the current K computer at the RIKEN Advanced Institute for Computational Science (AICS) in Japan.

At the International Supercomputing Conference (ISC’16) held in Frankfurt, Germany, at the end of June, Fujitsu announced that it will use 64-bit ARMv8 cores for the ‘post K’ supercomputer that it is building to replace the current K computer, housed at the RIKEN facility. The K computer is the current flagship supercomputer at RIKEN, ranking 5th on the latest Top500 list published in June 2016.

This move could help to drive the UK based ARM into mainstream HPC; this new supercomputer will likely be one of the top 10 most powerful supercomputers and by far the largest ARM cluster in the world. This new system will not only count as a vote of confidence towards ARM’s place in HPC, but it will also provide significant advances in RISC based application development and expertise.

During the ISC conference, Fujitsu released details of the new system during a presentation with Fujitsu vice president Toshiyuki Shimizu. Shimizu stated that the “post K” system, which is set to go live in 2020, will have 100 times more application performance than the K supercomputer.’

Shimizu also discussed the role of ARM and Fujitsu technology and how it will be used in the new system, saying “the Post K system will fully utilize Fujitsu’s proven supercomputer architecture.” He also stated that Fujitsu is acting as a “lead partner” for ARM HPC development and that the two companies were working to ‘realise an ARM powered supercomputer with high application performance.’

During the presentation, Shimizu described a roadmap for Fujitsu technology leading up to the development of the new RIKEN system. This roadmap includes includes the current generation of installed technology, including the current K computer and the Fujitsu’s PRIME HPC FX10, released in 2011 and the new PRIME HPC FX100 released in 2015.

RIKEN-logoThe PRIME HPC supercomputer platform has inherited many of the distinguishing features of the original K computer, including the RISC based SPARC64 bit processor and the Tofu 2 interconnect (Tofu2) which is integrated into the SPARC processor to increase node-to-node communication performance.

It is likely that the new RIKEN supercomputer will be based on the PRIME HPC FX100 platform, but it will be using the 64-bit ARMv8 processor as opposed to the SPARC XIfx processor currently used in the system.

ARM in HPC

The development of RISC-based HPC architectures stems from a desire to overcome the biggest obstacle to exascale computing – power efficiency.

Energy efficiency is a massive obstacle to increasing the size of supercomputers past 100 petaflops because it is becoming increasingly difficult to reduce power consumption of today’s processors much further. This is driving HPC developers to look to more innovative methods to further decrease energy consumption. From the use of passive cooling technologies to novel processor designs, such as embedded ARM processors, HPC users are putting considerable effort into creating more energy efficient supercomputers.

More information on the various types of processor available to the HPC industry can be found in the latest HPC Yearbook 2015/6.

The latest developments in cooling technology can be found in Scientific Computing World, as Tom Wilkie describes in his article from 2015.

Theoretically, with enough money and an infinite power budget it would be possible to build a 500 petaflop system or even a 1000 petaflop system today. However doing so would lead to poorly performing applications without considerable efforts to address applications scaling, and the total cost of the system would also be prohibitively expensive.

To address this fundamental hardware challenge facing exascale computing Fujitsu has chosen to look at the broader semiconductor market rather than try to adapt existing processors used in HPC today. That is not to say that RISC processors are new to the HPC industry. Fujitsu developed the previous K computer housed at RIKEN, which uses the RISC based SPARC64 VIIIfx 8C 2GHz processor.

However while Fujitsu was developing the SPARC processor line, ARM has had huge success in the embedded computing market, particularly in providing processors for mobile phones. Developing processors for these applications has led to ARM developing significant IP and expertise focused on creating processors that are hugely dependant on energy efficiency as they are often coupled with a small Lithium battery in a tablet or mobile phone.

Fujitsu and other ARM partners saw this requirement for the next class of HPC computing. Rather than trying to adapt current processor technologies – cutting back their power requirements – they looked to ARM, a company that has been developing these technologies for some years.

There is another added benefit because both the SPARC processor and ARM’s chips are RISC based, it should make the task of porting applications from the previous K computer onto this new ARM based supercomputer easier than if they were to port application s from RISC to x86 for example.

That is not to understate the importance of application. HPC experts at RIKEN have already begun the task of evaluating and preparing the most widely used applications for the arrival of this new ARM based system, and this process will likely continue past the point at which the new system has been delivered.

It remains to be seen whether this large ARM deployment will be enough to drive adoption of Arm processors across the HPC industry but to date, this is the strongest sign that ARM technologies will soon be used for large scale HPC.

If ARM and its partners can create an ecosystem of developers and application libraries specific to HPC, then this may be the start of new chapter in the race for exascale supercomputing.

This story appears here as part of a cross-publishing agreement with Scientific Computing World.

Sign up for our insideHPC Newsletter

Resource Links: