Sunny Austin was, for the first time, the host of this year’s Arm Research Summit and, just as in previous years, we saw a strong High Performance Computing (HPC) presence. Every session had an HPC track, and the rooms were packed. It has been a long journey since we started HPC in Arm, and it is refreshing to see the enthusiasm and engagement of both Arm and our partners in this space. In all, we took over a good chunk of the Summit!
The high performance activities kicked off on Sunday, with a workshop on the Impact of Arm hardware from an HPC application perspective (present and future looking). During the workshop, we heard from partners on next-generation systems, porting applications to the Scalable Vector Extension (SVE) architectural extension and current Arm platforms, user experiences and system usability requirements. We also heard about Arm’s historical HPC journey, from its humble mobile-oriented beginnings to the first petaflop Arm system on the Top500.
The main Summit parallel tracks kicked off with HPC Applications. These covered multiple talks on application optimization on Arm platforms - from oil and gas stencil computations, to mini-apps and solvers for computational materials and structural problems. We also had the pleasure of seeing a talk by Fujitsu on the performance evaluation of their AArch64 compatible A64FX.
Next was the HPC on Arm session, which covered users’ and integrators’ experiences with Arm machines. From prototypes to production, next steps, application optimization and early adoption. We heard from the University of Bristol, HPE, ANL and SNL.
HPC Applications II brought more talks, around embedding HPC at the edge. In future system requirements the shift is towards one trillion connected devices and the intensive data analysis requirements that this entails. We also heard about more porting experiences towards the Marvell ThunderX2 processor and enabling HPC applications for SVE, including FP16 usage.
During day 2, HPC users could choose from a few relevant talks, chief amongst which were the Arm HPC User’s Group and the Modeling session. As ever, the User Group drew in a large enthusiastic crowd who got to hear about new ecosystem engagements from Arm and Linaro. User experiences of moving towards SVE, future system requirements and heterogeneous directions. The Modeling session had a few talks by HPC users around full-system modeling for SVE, compilers and runtime tools by ORNL and RIKEN.
Finally, the Summit concluded with two BoF sessions, one on UK-Arm HPC: Accelerating the Use of HPC for Research and one on High Performance Graph Analytics: Algorithms, Programming, Architectures. Whilst the latter Is not as much traditional HPC, it is worth highlighting as it covered the large-scale problem of graph analytics, from software to hardware and architectures. This is one of the main “up and coming” high performance compute areas. The UK Arm HPC BoF covered the key areas of interest in the development of research Arm HPC technology, challenges and opportunities and how to co-ordinate the wider Arm ecosystem to help create a better Arm HPC environment.
As a true testament to the growth of the Arm ecosystem in this space, here is a list of all the institutions that gave HPC-related presentations, organized and ran events at the Arm Research Summit (and I hope I’ve not missed anyone!): CEA, Atos, Arm, KAUST, EPI, Jülich Supercomputing Center, GENCI, BSC, Old Dominion University, LANL, ORNL, Fujitsu, RIKEN R-CCS, University of Bristol, HPE, ANL, SNL, LBNL, NASA, Oxford InFoMM, Linaro, HLRS, LLNL, Georgia Tech, EPCC University of Edinburgh, SUSE, University of Leicester, PNNL, Tactical Computing Labs, University of Texas, Free University of Bozen.
Take a look at some of the key HPC presentations from the Arm HPC User's Group at the Arm Research Summit 2019.