Computer Facilities for the Development of Active LLM Evaluation Platform, 1 set

2026-11-03

Classification of the products to be procured : 1 Nature and quantity of the products to be purchased : Computer Facilities for the Development of Active LLM Evaluation Platform, set Type of the procurement : Purchase Basi requirements of the procurement : The computing facilities shall consist of an AI training computing system, an AI inference computing system, a general-purpose computing system, a control system, a large-capacity storage system, a high-speed network, as well as power supply and cooling facilities including uninterruptible power supplies (UPS) sufficient for stable system operation The AI training computing system shall be composed of multiple compute nodes. Each compute node shall be equipped with at least two CPUs and eight or more GPGPUs. The number of GPGPUs per compute node does not need to be uniform. The total number of GPGPUs across the entire system shall be at least 56 Each GPGPU installed in the compute nodes of the AI training computing system shall have at least 270GB of memory with ECC functionality. The memory bandwidth shall be at least 7.7TB/s, and the inter-GPU interconnect bandwidth shall be at least 1.8TB/s bidirectional The theoretical peak performance of a single GPGPU shall be at least 7 TFLOP for single-precision floating-point operations Each compute node of the AI training computing system shall be equipped with at least TB of main memory per CPU The AI inference computing system shall consist of multiple compute nodes specialized for inference. The architecture may be the same as of different from that of the compute nodes used in the AI training computing system The general-purpose computing system shall consist of multiple compute nodes, each equipped with at least two CPUs. The total number of CPUs across the entire system shall be at least 10 ⑦ Each compute node of the general-purpose computing system shall be equipped with at least 1TB of main memory per CPU ⑧ The control system shall consist of multiple control servers and provide the necessary functions for overall system management, including login nodes, batch processing servers, and other system-wide control functions ⑨ The large-capacity storage system shall provide an effective storage capacity of at least 100PiB and shall offer a high-performance, highly reliable parallel file system. The storage system shall be accessible at high speed from all servers ⑩ The high-speed network shall provide a bandwidth of at least 25Gbps and shall connect all servers and storage systems ⑪ The operating system shall be compliant with UNIX standards. The AI training computing system, AI inference computing system, and general-purpose computing system shall support batch processing functionality ⑫ The AI training and AI inference computing systems shall support programming languages such as FORTRAN, C, and C++, parallel programming environments such as MPI, development environments, and a variety of tools for AI training and AI inference ⑬ The facilities shall include power supply equipment, such as uninterruptible power supplies (UPS), and cooling systems. ⑭ A maintenance and support framework for both hardware and software shall be provided Time limit for the submission of the requested material : 1 : 0 2 April 202 Contact point for the notice : Kouji Tanaka, Common Infrastructure Group, General Planning Office, Universal Communication Research Institute, National Institute of Information and Communications Technology, 3- Hikaridai Seika-cho Soraku-gun Kyoto-fu 619-028 Japan TEL 0774-98-687 E-mail ci-post@khn.nict.go.jp

Japan

2026-03-26