Large-Scale Compute Nodes for the Integrated Infrastructure System for Simulation, Data, Learning, and Inferencing, 1 Set

2026-11-03

Classification of the products to be procured : 1 Nature and quantity of the products to be purchased : Large-Scale Compute Nodes for the Integrated Infrastructure System for Simulation, Data, Learning, and Inferencing, Set Type of the procurement : purchase Basi requirements of the procurement : A The system shall meet the following hardware requirements. B The existing radioactive sources must be removed and the new sources must be installed safely. a The computing system shall consist of compute nodes composed of accelerator nodes equipped with computational accelerators and general-purpose CPU nodes. b The total theoretical performance of the accelerator nodes (by double-precision floating-point arithmetics) shall be 160PFLOP or higher. Emulation is also acceptable. The CPUs of the general-purpose CPU nodes shall be of a 64-bit architecture. The general-purpose CPUs of the accelerator nodes may be of a different architecture. The CPUs shall support TEE (Trusted Execution Environment) functionality. d The total memory bandwidth of the accelerator nodes shall be 12PB/se or higher, and the total memory capacity shall be 220TB or higher. The accelerator devices should preferably support TEE (Trusted Execution Environment) functionality. e The total memory bandwidth of the general-purpose CPU nodes shall be 600TB/se or higher. f Each compute node shall be equipped with an NVMe SSD with 1.5TB or greater physical capacity. g The inter-node network interface of each compute node shall support a bandwidth of 800Gbps or higher per accelerator device on accelerator nodes. For general-purpose CPU nodes, the bandwidth shall be 800Gbps or higher per node. h Each general-purpose CPU node shall be capable of communicating directly with external systems with an aggregate bandwidth of 1,600Gbps or higher. Each accelerator node shall communicate with external systems via the general-purpose CPU nodes. i A highly reliable parallel file system with a storage capacity of 5.0PB or more shall be provided. It shall allow expansion to approximately 20PB without reconfiguration while maintaining the same configuration. It shall support data compression. The compute nodes shall be able to access the file system with a transfer performance of 1.0TB/se or higher. B The system shall meet the following software requirements. a The system shall operate on the Linux operating system. b The system shall provide compilers for Fortran 2008, C11, and C++2 or later for general-purpose CPUs, supporting automati SIMD vectorization and the OpenMP API version 4. or higher. For accelerator devices, it shall provide compilers for Fortran 2008, C11, and C++2 or later with automati parallelization and support for the OpenACC API version 2. or higher, the OpenMP API version 5. or higher, or both. A communication library supporting MPI version 3. or later shall be provided. d A Python runtime environment shall be provided. e Highly optimized numerical computation libraries and machine-learning libraries shall be provided. f A batch job system shall be provided. A single job shall be able to use both the general-purpose CPU node grou and the accelerator node grou simultaneously. g A container system shall be provided. Kubernetes shall be tightly integrated with the batch job system. C The total power consumption of the installed system, excluding cooling equipment, shall be 4.5MVA or less. Power capacity, cooling, and installation design shall ensure sufficient heat dissipation for continuous operation of CPUs, accelerator devices, memory, storage devices, and network switches. Direct water cooling shall be used for general-purpose CPUs, accelerator devices, and inter-node network switches. For other components such as storage, efforts shall be made to reduce operational costs through rear-door heat-exchanger water cooling or equivalent methods. The installation area, excluding cooling equipment, shall be 30 square meters or less. D The source replacement work must comply with applicable laws and safety requirements. E The system shall be operated using the management functions of the Integrated Infrastructure System for Simulation, Data, Learning, and Inferencing. Time limit for the submission of the requested material : 1 : 0 1 December, 202 Contact point for the notice : Accounting Team, Information Strategy Group, Information Systems Department, The University of Tokyo, 6-2- Kashiwanoha, Kashiwa-shi, Chiba-ken 277-0882, Japan, TEL 080-9452-382 All documents relating to the procurement should be written in Japanese.

Tokyo

2026-11-03