DTS-Led Consortium Completes Construction of RIKEN's Supercomputer 'ROQUO' and Begins Operations

DTS Co., Ltd. (Headquarters: Chuo-ku, Tokyo; President and CEO: Tomoaki Kitamura) has completed the system construction of the JHPC-quantum GPU supercomputer 'ROQUO (Rokko, hereinafter 'ROQUO')'—installed at the RIKEN Center for Computational Science (R-CCS) in Kobe City—developed under the initiative of RIKEN (Headquarters: Wako City, Saitama Prefecture; President: Makoto Gonokami, hereinafter 'RIKEN') to accelerate the integration of quantum computing and high-performance computing (HPC※1). The system has officially commenced operations as scheduled.

DTS formed a joint team comprising Digital Technology Corporation (Headquarters: Arakawa-ku, Tokyo; President and CEO: Toshiaki Kibe), a group company providing strong support across all project phases including equipment procurement, system construction, installation, and maintenance operations; ScaleWorX Inc. (Headquarters: Chiyoda-ku, Tokyo; President and CEO: Masahiko Yamada); and GIGA Computing Technology CO., LTD. (Headquarters: New Taipei City, Taiwan; CEO: Daniel Hou), each selected for their high level of expertise in their respective domains.

※1: HPC (High Performance Computing): High-performance computing

The 'ROQUO' system, constructed by the DTS-led consortium based on the requirements specified by RIKEN's Center for Computational Science (hereinafter 'R-CCS'), is among the world's leading production-grade systems utilizing NVIDIA's latest Grace Blackwell platform, the 'GB200 NVL4'. Additionally, it marks the first deployment in Japan of NVIDIA's high-speed networking technology 'InfiniBand XDR 800' and the NVIDIA Quantum-X800 Q3400 switches that form its fabric※2. This enables low-latency, high-bandwidth inter-node communication essential for quantum-HPC integration platforms.

※2: NVIDIA's latest Grace Blackwell platform integrates two NVIDIA Grace CPUs and four NVIDIA Blackwell GPUs per node via NVLink-C2C interconnect technology. While the GB200 NVL72 is optimized for large-scale generative AI training and inference with a 72-GPU configuration, the GB200 NVL4, with its 4-GPU configuration, is designed to support HPC (scientific computing) applications, offering superior flexibility and cost-performance balance in deployment and operation. Both 'ROQUO' and the supercomputer 'RIKYU', announced on the same day, represent world-leading production systems based on the GB200 NVL4 platform.

During preparatory operations, the High Performance Linpack (HPL) benchmark test achieved 19.8 petaflops (PFLOPS; P for peta equals 1 quadrillion) in double-precision floating-point operations (FP64)※3. Furthermore, by employing warm-water-cooled servers and 'free cooling' technology that leverages ambient air, the system cools water using cooling towers alone—without compressors—significantly reducing energy consumption for cooling. 'ROQUO' uses 32°C water as coolant, which can be naturally cooled even during Kobe's hottest summer months, achieving over 20% reduction in total power consumption compared to similar-scale supercomputers.

※3: HPL is a benchmark that measures processing performance for solving dense linear systems of equations and is widely used as the evaluation metric for the TOP500 global supercomputer performance ranking. The measured value cited in this press release was obtained using all computing nodes of 'ROQUO' (135 compute nodes, 540 GPUs).

'ROQUO' is named after Rokko Mountain (Rokko-san), a landmark symbolizing Kobe City, and serves as a new computational foundation for quantum-HPC integration platforms. The front panel features a design inspired by the night view of Kobe as seen from Rokko Mountain, with logos of the participating companies displayed.

This operation is part of the JHPC-quantum project, funded by NEDO (New Energy and Industrial Technology Development Organization) under the 'Post-5G Information and Communication Systems Infrastructure Strengthening R&D Program (JPNP20017)', specifically the commissioned project 'Research and Development of a Quantum-Supercomputer Collaborative Platform for Expanding Computable Domains' (Project Leader: Dr. Mitsuhisa Sato, Director, Quantum-HPC Integration Platform Division, R-CCS, RIKEN).

Overall Architecture of the Quantum-HPC Integration Platform

R-CCS (Kobe) hosts 'Fugaku', 'ibm_kobe', and 'ROQUO' (labeled 'GPU cluster' in the diagram), interconnected via SINET with 'Kei' at RIKEN (Wako), Osaka University, the University of Tokyo, and SoftBank.

Company names, product names, and logos mentioned are trademarks or registered trademarks of their respective companies.

(Reference)

RIKEN Press Release, June 19, 2026

https://www.riken.jp/pr/news/2026/20260619_2/index.html

Digital Technology Corporation Press Release, June 19, 2026

https://www.dtc.co.jp/press_20260619

<Overview of DTS Co., Ltd.>

DTS is a total systems integrator (Total SIer) with comprehensive capabilities. It provides end-to-end services—from consulting and system design/development to infrastructure construction and operations—primarily for the finance, information and communications, manufacturing, public, and construction sectors.

Additionally, the DTS Group delivers high-value-added services by leveraging diverse technical expertise in system-related fields. https://www.dts.co.jp/

Headquarters: Empire Building, 2-23-1 Hachobori, Chuo-ku, Tokyo 104-0032