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.
| Oakforest-PACS | |
|---|---|
| Name | Oakforest-PACS |
| Location | JAEA/Tokyo Metropolitan University/Kyoto University |
| Architecture | Fujitsu PRIMEHPC FX10, Intel Xeon Phi |
| Cpu | Intel Xeon Phi 7220 (KNL) |
| Memory | 16 GB MCDRAM per node (configurable) |
| Peak | 25.1 PFLOPS (theoretical) |
| Date deployed | 2017 |
| Operators | Joint Center for Advanced High Performance Computing |
Oakforest-PACS
Oakforest-PACS is a Japanese high-performance computing system deployed as a national supercomputing resource. The system serves research communities across Japan and international collaborators, supporting computational projects tied to institutions such as University of Tokyo, Kyoto University, Tokyo Institute of Technology, and research programs funded by the Ministry of Education, Culture, Sports, Science and Technology (Japan). Oakforest-PACS provides a platform for simulation and data analysis in fields connected to centers like the RIKEN and agencies like the Japan Aerospace Exploration Agency.
Oakforest-PACS was commissioned to address needs in large-scale simulation workloads typical of projects at KEK, JAXA, National Institute of Advanced Industrial Science and Technology, and university consortia including Osaka University and Tohoku University. Designed to bridge between national facilities such as the Fugaku successor programs and regional clusters like the HPCI, Oakforest-PACS offers resources for research in disciplines linked to teams at RIKEN Center for Computational Science, The University of Tokyo Institute of Medical Science, and departments affiliated with the Japan Science and Technology Agency. Collaborative users include groups that have links with international partners such as Lawrence Livermore National Laboratory, Argonne National Laboratory, and CERN.
The system architecture is built around the Intel Xeon Phi Knights Landing (KNL) many-core processors integrated in Fujitsu cabinets similar to designs by Fujitsu for PRIMEHPC systems. Each compute node combines KNL processors with high-bandwidth MCDRAM and DDR4 memory configurations used in workflows from centers like Sandia National Laboratories and Los Alamos National Laboratory. Oakforest-PACS incorporates high-speed interconnects influenced by technologies from InfiniBand vendors and follows rack and cooling layouts similar to deployments at National Supercomputing Center in Wuxi and Pawsey Supercomputing Centre. System management borrows provisioning practices used at NERSC and PRACE member centers.
Oakforest-PACS achieved notable benchmark results in workloads comparable to those run on machines like K computer and Fugaku predecessor systems. Measured performance for dense linear algebra and fluid dynamics codes drew comparisons to results published by groups at Stanford University, MIT, and ETH Zurich. Benchmarks such as LINPACK-like tests and domain-specific scaling studies were performed by teams associated with Riken AICS and academic groups from Kyushu University. The system sustained high utilization for parallel applications developed at centers like NASA Ames Research Center and European Centre for Medium-Range Weather Forecasts.
The software stack for Oakforest-PACS supports programming models widely used in computational research, including MPI implementations from projects like Open MPI and MPICH, threading libraries such as OpenMP, and accelerators programming approaches influenced by OpenACC and Intel Parallel Studio. Compilers and runtime systems from Intel and ecosystem projects used in labs like CERFACS and NERSC enabled performance tuning. Users ported applications from repositories maintained by teams at Argonne National Laboratory and Lawrence Berkeley National Laboratory, employing tools like PETSc, Trilinos, HDF5, and profiling suites similar to those at PRACE.
Operational responsibilities rested with a consortium including Tokyo Metropolitan University, University of Tsukuba, and administrative frameworks connected to Japan Agency for Marine-Earth Science and Technology. Scheduling and resource allocation adopted practices used at national facilities such as HPC Wales and workflow portals similar to those at XSEDE. Maintenance cycles mirrored experiences from centers like SDSC and HLRS, with software support coordinated alongside vendors including Fujitsu and hardware service teams from Intel.
Research applications included climate modeling linked to groups at Meteorological Research Institute (Japan), computational fluid dynamics from laboratories at Nagoya University, seismic simulations coordinated with National Research Institute for Earth Science and Disaster Resilience, and biomedical modeling in collaboration with Osaka University Medical School. User communities also included computational chemistry groups at Kyoto University and astrophysics teams affiliated with The University of Tokyo Graduate School of Science and international consortia such as Square Kilometre Array development teams. Cross-disciplinary projects drew participation from researchers at Tohoku University, Hokkaido University, and institutes collaborating with European Southern Observatory.
Oakforest-PACS was developed amid national initiatives to expand Japan's high-performance computing capacity, with contributions from vendors and institutions that previously participated in projects like K computer and national programs under the Council for Science, Technology and Innovation (Japan). Its procurement and deployment followed competitive processes influenced by procurement models seen in procurements involving Fujitsu and collaborations with academic partners such as Nagoya University and Keio University. The system's operational life intersected with planning cycles for successor facilities including national efforts coordinated through the High-Performance Computing Infrastructure (HPCI) and partnerships with research centers like RIKEN and JAXA.