This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| Fujitsu A64FX | |
|---|---|
| Name | A64FX |
| Designer | Fujitsu |
| Architecture | ARMv8.2-A Scalable Vector Extension |
| Cores | 48+2 (FCU) |
| Process | 7 nm |
| L1cache | 48 KB I$ / 64 KB D$ per core |
| L2cache | 8 MB per core complex |
| Memory | HBM2 |
| Launch | 2019 |
| Used in | Fugaku, PRIMEHPC FX700 |
Fujitsu A64FX is a 64-bit ARM-based scalar/vector processor designed for high-performance computing, combining ARMv8.2-A features with Scalable Vector Extension (SVE) and high-bandwidth HBM2 memory to target exascale-class workloads. It serves as the central CPU of notable supercomputer systems and was developed through collaboration between industry and research institutions to optimize scientific simulation, data analytics, and machine learning. The design emphasizes energy efficiency, vector throughput, and memory bandwidth to compete with traditional x86 and accelerator-based architectures.
The A64FX project was announced amid global exascale efforts involving organizations such as Riken, Fujitsu Limited, Arm Holdings, Taiwan Semiconductor Manufacturing Company, European Commission, and national laboratories including Los Alamos National Laboratory and Lawrence Livermore National Laboratory. It targets workloads popularized by benchmarks like High Performance Linpack, Graph 500, SPEC CPU, and scientific codes from communities around Weather Research and Forecasting Model, GROMACS, LAMMPS, and Quantum ESPRESSO. The processor participates in international comparisons alongside systems from IBM, Intel Corporation, NVIDIA, AMD, and proposals from Cray Research and HPE. It contributed to recognition in awards akin to the TOP500 milestones and drew interest from supercomputing centers at institutions such as Oak Ridge National Laboratory and Argonne National Laboratory.
A64FX implements the ARMv8-A instruction set family with extensions from Scalable Vector Extension to deliver 512-bit or variable-width vector operations, aligning with standards from ARM Ltd. and influenced by research from University of Tokyo, Tohoku University, and RIKEN Center for Computational Science. The core complex features 48 compute cores and 2 assistant cores for operating-system tasks, connected via a mesh and coherent fabric reminiscent of designs from Sun Microsystems and Intel Xeon Phi initiatives. It integrates HBM2 stacks provided through packaging partnerships with TSMC and memory IP contributors linked to SK hynix and Micron Technology. On-chip features include hardware support for double-precision floating point for codes developed at groups like CERN, Max Planck Society, and Los Alamos National Laboratory.
Designed for high memory bandwidth and low power per FLOP, A64FX achieved notable results on benchmarks such as High Performance Linpack and graph analytics measured by Graph 500, often compared with systems powered by NVIDIA A100, Intel Xeon, and AMD EPYC. Performance gains were documented in simulations from climate centers like Meteorological Research Institute (Japan) and chemistry codes used at Riken and RIKEN AICS. Its vector engine accelerated workloads typical in genomics pipelines used at Broad Institute and machine learning training similar to tasks run on clusters at Google and Amazon Web Services, while maintaining thermal and energy profiles studied by researchers at Keio University and Tokyo Institute of Technology.
The A64FX is the heart of the flagship system developed in collaboration with RIKEN, deployed as a full-scale supercomputer by centers such as Riken Center for Computational Science for scientific applications including climate modeling, bioinformatics, and materials science. Commercial offerings by Fujitsu Limited include cabinets and clusters branded under models like PRIMEHPC FX700, used by institutions comparable to Japan Agency for Marine-Earth Science and Technology and corporations such as Fujitsu Limited's enterprise customers. The processor has been integrated into testbeds and pilot installations at universities including University of Tokyo and national labs in Japan and abroad.
The software stack for A64FX spans operating systems and toolchains from projects like Linux, OpenHPC, and compilers including GNU Compiler Collection, LLVM, Fujitsu Compiler, and vendor-tuned libraries analogous to Intel MKL and NVIDIA cuDNN. Parallel programming models supported include MPI, OpenMP, and efforts to optimize MPI implementations from groups at Argonne National Laboratory and Lawrence Berkeley National Laboratory. Scientific packages were ported and optimized by teams at institutions such as RIKEN, University of Tsukuba, National Institute of Advanced Industrial Science and Technology, and collaborations with HPC centers around the world. Performance libraries and profilers used include tools similar to VTune, TAU, and HPCToolkit.
Development involved collaborations between corporate and academic partners, with silicon fabrication using industry-leading 7 nm processes from TSMC and packaging performed in coordination with partners in Japan and Taiwan. The processor's creation drew on prior designs from companies like Fujitsu Microelectronics and research from national programs such as Japan's Society 5.0 initiatives and projects funded by agencies like METI and JSPS. Manufacturing supply chains connected to semiconductor equipment vendors like ASML and materials suppliers akin to Sumco and Shin-Etsu Chemical.
The processor was praised by international benchmarking bodies and academic reviewers for energy-efficient design and enabling a non-x86 route to exascale-class performance, prompting discussion among stakeholders such as TOP500 organizers, supercomputing centers at Oak Ridge National Laboratory and Los Alamos National Laboratory, and vendors like HPE and Cray. It influenced procurement strategies in national research infrastructures, inspired academic studies at University of Cambridge and MIT, and spurred ecosystem investments from cloud providers comparable to Amazon Web Services and Microsoft Azure exploring ARM-based HPC offerings. The A64FX's deployment contributed to policy and industrial dialogues involving Japanese government research initiatives and international collaborations across science and engineering.