A great deal of news that is arrived during the Supercomputing Conference 2019 (SC19) in Denver, Colorado last month. Every year at the SC, we highlight the work and growth of the Arm HPC ecosystem. The focus is on our various partners in this space. At SC19, we saw several significant milestones announced. Below, I have tried to fit the news into one of two categories: Arm HPC Software Ecosystem and Arm Hardware Ecosystem. I have also included a brief statement about why each is important to Arm.
While much of the news at SC19 was Arm hardware partner related, there was a significant software item to cover right up front. Kicking off the week with a GPU bang, NVIDIA announced the availability of their CUDA® toolkit for Arm and a partner reference design for building GPU-accelerated Arm servers. At NVIDIA and Arm, we have two powerhouse companies that know how to build ecosystems and collaborate with strategic partners. This is all for the benefit of the broad compute market whether it is supercomputing datacenters or cloud providers, all the way out to the edge. Customers who are interested in deploying Arm for HPC have a choice of a GPU model or a CPU-centric model, and we expect to see partners such as Marvell, Huawei, and Ampere touting support for NVIDIA V100 GPUs in their platforms.
Jensen Huang, Darren Cepulis, and Nick Forrington at SC19
Another highlight of the SC19 week was our Quarterly Catalyst UK FTF meeting. Also, there is news that the Catalyst UK collaboration won the HPCWire Reader’s Choice Award for Best HPC Collaboration. This collaboration brings together both hardware and software ecosystem partners. The idea is to drive forward the maturity of the Arm HPC software ecosystem and to run real scientific simulations on Arm at three major UK academic sites (Bristol, EPCC, and Leicester).
Rounding out the highlights of SC19 in the HPC software category, the OpenHPC project announced its latest release (v1.3.9) in conjunction with the conference. Arm is a long-standing member of this organization which provides a baseline set of open-source HPC software packages. These are built to deploy on various standard platforms and Operating Systems (for example, CENTOS7 and SLES12), including those based on the Arm architecture.
While Arm has maintained a focus on strategic HPC customers, which rely heavily on open-source and owned codes. We recognize the need for further enablement of key ISVs for sites that support a broader cross-section of users and applications. With that in mind, I also wanted to highlight the great work Altair has been doing in porting and validating their codes on Arm at scale. Eric Lequinou, VP of Development at Altair, presented at the Arm HPC User Group covering his work running and benchmarking Altair RADIOSS on Arm. There was extra Altair on Arm content that is shown in both the Altair and Arm booths. This highlights how important our end-users and customers can influence partner ISVs to provide commercial support for their important applications on Arm.
Moving to the HPC hardware section, SC19 was a banner year for Arm, both literally and figuratively. This year, our strategic HPC partner Fujitsu was front and center with several significant announcements around their Arm-based A64FX chip and their efforts to deploy. To celebrate this important milestone, Arm’s Brent Gorda was invited to take part in a special Sake toast at the Fujitsu booth during the opening evening of the event.
Brent Gorda and Toshinori Kujiraoka, with Andy Wafaa at SC19
Fujitsu announced ahead of SC that they have begun deployment of the Fugaku supercomputer system at RIKEN in Kobe, Japan. This system is important in that it may be the first exascale-class system in production by this time next year. It boasts nearly 8m cores based on Arm and features our Scalable Vector Extensions (SVE) technology.
By SC19, Fujitsu had already built up the first two functional racks and submitted their benchmarks to the Top500 and Green500 lists with significant results. They came in at #1 on the Green500 list, beating out dozens of GPU accelerated systems with their CPU-centric architecture. This proof-point has long been touted as a key advantage of using the Arm architecture and it is great to see it come to fruition. Fujistu also landed at a respectable #159 on the Top500 list, garnering 2PFlops of performance using only two racks of servers. While the Arm-based HPE Astra supercomputer held strong at #196. If there was a compute density Top 500 list, Fujitsu would have likely been #1 on that too.
Fujitsu banner showing their newest Arm-based Supercomputer line at SC19 in Denver.
Besides supplying servers with A64FX chips in racks to RIKEN, Fujitsu also announced their intent to sell them world-wide as they launched the "Fujitsu Supercomputer PRIMEHPC" series. This is with the liquid cooled FX1000 and the air-cooled FX700 models targeting both strategic and mainstream HPC markets.
Cray and Fujitsu also announced a strategic partnership with Cray. They will market and support supercomputer solutions based on the A64FX chip in the Exascale era. Broadening its choice of Arm processor technology, the Cray CS500 system can target for HPC and AI workloads with industry-leading efficiency and performance. Interest was immediate from various strategic sites such as LANL and ORNL. And academic sites such as Stony Brook University in the USA and Bristol University in the UK.
Still flying a bit under the radar is the EU strategic initiative to design and market their own processor technology. The European processor Initiative announced its official roadmap with Arm technology as the choice for its host processor cores. We are excited to see that they are leveraging our next generation “Zeus” core and expect great future performance for HPC applications.
Finally, another key item that is important to Arm (and its ecosystem partners) is the proposed expansion and spin-off of the Arm HPC User Group (AHUG) into a strategic end-user lead organization. Arm has always collaborated heavily with our strategic users. Whether in a research capacity or an ecosystem enablement capacity, we have worked as a hub of information for our partners and end users. We continue to see strong interest which enables the community to take greater control of information sharing and collaboration. You can expect to see more AHUG events to happening 2020 under this evolving framework.
Please stay tuned and keep an eye on our community AHUG pages or the Arm Developer pages for more info as it arrives.
Please join in the action and planning as it kicks off.