A performance analysis of the first generation of HPC‐optimized Arm processors

Abstract

Summary In this paper, we present performance results from Isambard, the first production supercomputer to be based on Arm CPUs that have been optimized specifically for HPC. Isambard is the first Cray XC50 “Scout” system, combining Cavium ThunderX2 Arm‐based CPUs with Cray's Aries interconnect. The full Isambard system will be delivered in the summer of 2018, when it will contain over 10 000 Arm cores. In this work, we present node‐level performance results from eight early‐access nodes that were upgraded to B0 beta silicon in March 2018. We present node‐level benchmark results comparing ThunderX2 with mainstream CPUs, including Intel Skylake and Broadwell, as well as Xeon Phi. We focus on a range of applications and mini‐apps important to the UK national HPC service, ARCHER, as well as to the Isambard project partners and the wider HPC community. We also compare performance across three major software toolchains available for Arm: Cray's CCE, Arm's version of Clang/Flang/LLVM, and GNU.

Benchmarking the first generation of production quality Arm‐based supercomputers

“…To evaluate the state of the software ecosystem for Arm, we used all three compiler families available to us: GCC, the LLVM‐based Arm HPC Compiler, and Cray's CCE. The Isambard node‐level performance paper at CUG 2018 was the first study to compare all three of these compilers targeting Arm . Likewise, for the Intel processors, we used GCC, the Intel compilers, and Cray's CCE.…”

“…The Isambard system represents a collaboration between the GW4 Alliance (the universities of Bristol, Bath, Cardiff, and Exeter), the UK's Met Office, Cray, Arm, and Marvell, with funding coming from EPSRC. At CUG, we showed initial single‐node results based on early‐access systems that demonstrated that the performance of the Arm‐based ThunderX2 CPUs was comparable to state‐of‐the‐art x86 processors available at the time. The full Isambard system was delivered in November 2018, and we presented our initial scaling results at CUG, showing that Arm‐based supercomputers remain competitive at scale, running real scientific workloads.…”

Special Issue of the Cray User Group (CUG 2018)

“…A subset of papers in this Special Issue talks about the advances made in the Cray architecture, operations, and management, a very relevant topic to a large majority of sites that operate Cray supercomputers. Another subset of the papers touched on the latest developments and techniques in performance analysis and benchmarking in supercomputing …”

“…However, given the cost per FLOP ratio that ARM processors offer, they are very compelling. McIntosh‐Smith et al presented the performance results of what is likely the first Cray supercomputer based on ARM central processing units. This machine is located at the GW4 Alliance in the United Kingdom and is named “Isambard.” The Cray XC50 “Scout” form factor–based machine features Cavium ThunderX2 ARM 64‐bit central processing units and the Aries interconnect.…”

Design trade-offs for emerging HPC processors based on mobile market technology

“…In fact, these trends are already materializing and vendors have started to roll out competitive processors that target the HPC market based on lowpower ARM-based systems-on-chip (SoC), e.g., the ThunderX processors family [3]. In addition, several HPC companies are already deploying large ARMbased machines, e.g., Cray with the deployment of a 10,000+ ARMv8-A core system [4,5], and in the near future Fujitsu's next flag-ship machine -The Post-K [6].…”