A few people have asked about my new role at Arm and, more directly, about the attractiveness of Arm to the HPC community. I am quick to answer enthusiastically that I view Arm as disruptive to HPC and am happy to have a role here. Today, Hyperion Research values the HPC server market at $12 billion, but that grows substantially to $19 billion by 2022, and I expect Arm to be a significant force in this time frame, providing architectural options to the community.
I've been fortunate to have been involved in a number of disruptive changes to HPC over my career. I was part of the team at the Lawrence Livermore National Laboratory (LLNL) pioneering microprocessor based parallel computing in the Massively Parallel Computing Initiative. Also at LLNL, I held the contracts with IBM for the BlueGene architecture. I founded and was CEO of Whamcloud, a startup focused on the Lustre filesystem. After selling to Intel, I was aligned tightly with the HPC business there as the General Manager for the High Performance Data Division.
Arm's business model is interesting and enabling for HPC. We create and license processor IP so others can take that IP and create solutions in silicon (Cavium/Marvell, Fujitsu, Ampere and others) that is then taken to the HPC market place (HP, Cray, ATOS/Bull, and many more). Each of these players can optimize the solution as they see fit, resulting in a diverse set of offerings for the market.
Arm, along with others (Cray, Linaro, HP, Cavium, Fujitsu) work together to provide the software necessary for these solutions. The compatibility of the ISA provides a foundation upon which software is compatible across the platforms.
Since arriving at Arm, I've been quite impressed with how mature the software ecosystem is. Arm supports both commercial and open-source toolchains and has enabled the porting of hundreds of foundational HPC packages and applications to the Arm ISA. We also have a strong team of HPC experts on hand to engage with strategic end-users in support of their deployments and ecosystem requirements.
As for indication of disruption, already we know of exascale designs based on Arm to be delivered in the future. Today, there are a number of Arm-based systems being installed internationally. These are primarily based on Marvell’s ThunderX2 and early results indicate market-leading performance, due to their large number of cores per socket and the strong memory bandwidth. Looking to the future and Arm's cool, new SVE (vector) instructions. On that subject, Fujitsu announced the A64FX with SVE implementation at HotChips this summer, causing quite a bit of chatter in the community. I’ve joined with others following @ProfMatsuoka to be an early recipient of information on the chip and the deployment at RIKEN in Japan.
I’m looking forward to my new role having a big impact on HPC. You can follow our progress on our dedicated HPC Ecosystem Developer page.
Brent Gorda, Sr. Directory for HPC Business.
[CTAToken URL = "https://sc18.supercomputing.org/" target="_blank" text="Want to learn more? Join us at SuperComputing '18 in November" class ="green"]
Happy that things are shaping up for ARM in HPC but I look forward to seeing more happening in the HPEC arena too.