⑴ Classification of the products to be procured : 14

⑵ Nature and quantity of the products to be purchased : Interdisciplinary Large-Scale Computing System 1 unit

⑶ Type of the procurement : Rent

⑷ Schedule time of introduction : On or after April 2024

⑸ Basic requirements of the procurement :

A Supercomputer system

The system to be procured consists of a computing subsystem, a storage system, and other hardware and software components for the stable and efficient operation of the total system.

① Computing subsystem

A-1 The computing subsystem must be a cluster type parallel computer composed of multiple computer nodes. Hereafter, the computer node, or the node in short, is defined as a computer system composed of multiple CPU cores and a main memory shared by them. Some of the nodes must be equipped with GPU (Graphics Processing Unit). The node having GPUs is called a GPU node, and the node which is not equipped with GPUs is called a CPU node.

A-2 Each node must have multiple multi-core processors and a main memory of 512 GiByte or larger. The theoretical peak performance on double-precision floating point in a CPU node must be 5 TFLOPS or higher, and that in a GPU node must be 240 TFLOPS or higher.

A-3 The aggregate theoretical peak bandwidth of the main memory access in each node must be 600 GByte/sec or higher.

A-4 Each node must be equipped with SSDs, or secondary storages that have the same capabilities, of 300 GiByte or larger in total.

A-5 The total sum of the theoretical peak performance on double-precision floating point in all nodes of the computing subsystem must be 30 PFLOPS or higher, and the total main memory size must be 375 TiByte or larger.

A-6 The theoretical peak injection bandwidth of each node to inter-node interconnect must be 50 GByte/sec or higher in a CPU node and 100 GByte/sec or higher in a GPU node. The inter-node interconnection among all CPU nodes must have full bi-section bandwidth, and that among all GPU nodes must have full bi-section bandwidth. The bi-section bandwidth when partitioning the nodes into CPU nodes and GPU nodes must be 4 TByte/sec or higher.

A-7 Each node must have hardware management mechanisms and software that monitors the operational state.

A-8 The computing subsystem must operate with peak load consuming 1.2 MW electric power or less.

A-9 The computing subsystem must have the software stack shown in Item ②.

② Software stack

A-10 Automatic parallelizing compilers for Fortran, C, and C++ must be provided. These compilers must support OpenMP version 5.0 or later.

A-11 Java and Python platform must be provided.

A-12 High-performance parallel numerical libraries, program development environment and debuggers must be provided.

A-13 Parallel programming library that supports MPI version 3.1 or later must be provided.

A-14 UNIX operating system must be provided.

A-15 System software for both interactive and batch processing modes must be provided.

A-16 The same operations as in the current system, namely the following two types of operations, must be provided : (ⅰ) the conventional metered operation, (ⅱ) the group occupation operation in which a specific user group is allowed to occupy a certain amount of computational resources.

③ Storage system

A-17 The total capacity of disk arrays must be 32 PByte or larger.

A-18 The aggregate data transfer bandwidth between the storage system and the computing subsystem must be 500 GByte/sec or higher.

A-19 The storage system must operate with peak load consuming 120 kW electric power or less.

④ Common requirements

A-20 Porting and tuning supports for main application programs running on the current system must be provided.

A-21 A wide spectrum of application software and various free software must be installed.

A-22 User management functions for accounting, collecting usage statistics, and authentications must be provided.

A-23 The supercomputer system must be connected to the campus network of Hokkaido university.

B Cloud system

The cloud system consists of the server system and the storage system both installed on campus and the remotely installed archive system.

B-1 As the server system installed on campus, it must have 5,000 or a higher number of cores in total, which operates at 2 GHz or higher rated frequency, and 15 TByte or larger aggregated

main memory, whose individual transfer bandwidth is 4,800 MT/s or higher. The accompanying server boot storage is required to offer 250 TByte or larger physical capacity and high access performance that enables the servers to boot in less than 15 minutes.

B-2 The server system must offer connectivity to the existing campus network and SINET6 with a 100 Gbps class high-speed network s. The internal network should also offer a 100 Gbps class transfers and have a virtual network function that can separate networks from others for each user tenant.

B-3 As the storage system, it must be a storage system with 2 PByte or larger physical capacity, high access performance, and high reliability, and offer online storage service that can be accessed from many user devices via the Web or dedicated client software.

B-4 The archive system must be installed on a remote site and have 5 PByte or larger archival capacity. The existing data must be migrated from the current remote archive system.

B-5 The software environment of the server system must be realized by either of a containerized platform offered by Kubernetes, an on-premises cloud platform d on virtual machines, or a hyper-converged system.

B-6 The server system must offer security and privacy protections d on the technologies of memory data encryption offered by processors, storage data encryption, and network communication encryption.

B-7 The interoperability with existing public cloud service providers should be considered.

C Others

C-1 Management systems must be provided to operate the supercomputer and the cloud system smoothly.

C-2 The system must possess high reliability for both hardware and software to provide stable services for 24 hours/365 days basis.

C-3 The available electric power for the computers is totally 2 MVA, which is provided by two 0.75 MVA transformers and a 0.5 MVA one. In addition, a 0.75 MVA transformer is available for the cooling system. The system must be designed to minimize the total electricity consumption, installation space, and operational costs. Also, automated system management and sufficient operation support must be provided.

C-4 Support system must be necessary for both hardware and software to recover from systems failures and security incidents.

C-5 Users' assets such as data, files and software, and operational management systems including servers and their network configurations must be migrated from the current system.

⑹ Time limit for the submission of the requested material : 17 : 00 12 September, 2022

⑺ Contact point for the notice : KAMIOKA Junji, Procurement Section, Procurement Division, Hokkaido University, Kita 8 Nishi 5 Kita-ku Sapporo-shi 060-0808 Japan, TEL 011-706-4552