Quantum-AI Hybrid Computing Infrastructure System A 1 Set

2026-11-03

Classification of the products to be procured : 71, 2 Nature and quantity of the products to be purchased : Quantum-AI Hybrid Computing Infrastructure System A Set Type of the procurement : purchase Basi requirements of the procurement : ABCI-Q System A consists of hardware comprising a high-performance computing system, various networks, and other related components ; power supply and cooling facilitiessufficientforoperatingthesystem ; and software necessary to maximize utilization of the system. The followings summarizes the requirements for ABCI-Q System A : A.The High-Performance Computing System shall provide a theoretical inference performance of at least 3. EFLOP and a theoretical peak training performance of at least 0.2 EFLOPS. B.The High-Performance Computing System shall include high-speed memory integrated with the accelerator processors. The total memory capacity shall be at least 2 TB, and the total peak memory bandwidth shall be at least 1. PB/s. C.The Computational Network shall connect the High-Performance Computing System and the Large Capacity Storage System (DDN EXA Scaler) of the existing ABCI-Q System H (hereinafter referred to as the "System H Large-Scale Storage System"). In addition, the network must employ links with bandwidth of at least 80 Gbps to achieve the highest possible bisection bandwidth. The theoretical bandwidth between compute nodes, as well as between compute nodes and the System H Large-Scale Storage System, must be more than 80 Gbps. D.The Service Network shall connect devices requiring external access to the High-Performance Computing System and shall also be connected to the computational network, the management network, and the inter-site connection network. E.The system shall be designed to efficiently utilize the installation space, power supplies, and colling facilities provided by AIST. The effective total power consumption of the system shall not exceed 300kW. F.The system shall include software stacks that maximize system utilization and improve the efficiency and performance of both batch-processing workloads, such as large-scale AI model development and quantum circuit simulation, and online services such as AI agent inference and serving. Time-limit for the submission of the requested material : 1 : 00, November 13, 202 Contact point for the notice : IRIE and HARUTA, Procurement Division, National Institute of Advanced Industrial Science and Technology, 1-1- Higashi, Tsukuba, Ibaraki, 305-856 Japan. TEL 050-3522-414 or 050-3522-5845

Japan

2026-10-02