Supercomputing Division, Information Technology Center, The University Tokyo

JCAHPC decided to introduce "OFP-II"


  • JCAHPC decided to introduce a new supercomputer system "OFP-II" with a total peak performance of 79.5 PFLOPS.
  • OFP-II will be jointly procured and operated by the University of Tokyo and the University of Tsukuba.
  • OFP-II will be the first large-scale system with NVIDIA GH200 Grace Hopper Superchip
  • Operation will be started in January 2025.


 In March 2013, Center for Computational Sciences, University of Tsukuba (CCS), and Information Technology Center, the University of Tokyo (ITC) agreed to establish the Joint Center for Advanced High-Performance Computing (JCAHPC). JCAHPC is an organization to build and operate a stat-of-the-art, large-scale, high-preformance computing infrastructure by introducing supercomputer systems to be jointly procured, and has aimed to provide large-scale, ultra-high-speed computing capabilities to promote cutting-edge computational science and to contribute to the promotion of academics and science and technology in Japan. As a result, Oakforest-PACS started operation in October 2016 and ended in March 2022 after 5 years and 6 months of operation.
 JCAHPC has been considering the design, construction, and operation of the successor to the Oakforest-PACS since November 2019. JCAHPC decided to introduce a new supercomputer system "OFP-II" with a total peak performance of 79.5 PFLOPS. The operation is scheduled to start in January 2025. Inheriting the philosophy of the introduction of Oakforest-PACS, OFP-II aims to promote novel computational science methods leveraging AI, such as AI for HPC/Science, to offer a platform supporting Society 5.0 by integration of simulation, data analysis, and machine learning, in addition to support users of large-scale applications.

Specification of OFP-II system

  • Theoretical Peak Performance: 79.5 PFLOPS
  • OFP-II has General-purpose CPU node adnd Accelerator node
    • General-purpose CPU node: 190 nodes with two Intel Xeon CPU Max 9480 per node
    • Accelerator node: 1,120 nodes with NVIDIA GH200 Grace Hopper Superchip per node
  • Shared File System: NVMe SSD 10.3 PB

Overview of OFP-II


OFP-II Specification

OFP-II Specification

Road to OFP-II

Road to OFP-II