Fujitsu Accelerates AI Research and Development, Demonstration, and Social Implementation Through Industry-Government-Academia Collaboration
Fujitsu has received an order for “AI Bridging Cloud Infrastructure” from National Institute of Advanced Industrial Science and Technology to expand AI Research and Development in Tokyo.
The “AI Bridging Cloud Infrastructure” is a 2-way rack type (4U) server “FUJITSU Server” that is equipped with 8 GPUs per node and is suitable for large-scale analysis work that utilizes AI such as deep learning. It consists of 120 units of the next model of “PRIMERGY GX2570” and a large-capacity storage system of 11.2 petabytes.
The theoretical peak performance of half-precision floating-point arithmetic, which is regarded as important in the AI field, is 300 petaflops, and double-precision floating-point arithmetic, which is mainly used in the conventional simulation field, is 19.3 petaflops. It is expected to be.
The “AI Bridging Cloud Infrastructure” will be added to the AI data center building (location: Kashiwa City, Chiba Prefecture) of the Kashiwa Center of the National Institute of Advanced Industrial Science and Technology, and is scheduled to start operation in 2021.
AIST is an “AI bridging cloud infrastructure that enables advanced AI processing by coordinating algorithms, big data, and computing power in order to accelerate cutting-edge research and development of AI technology and social implementation in industrial fields. (Hereinafter, “ABCI”) “has been in operation since August 1, 2018. “ABCI”, which we received and built, was a cloud platform for AI applications with the world’s highest level of machine learning processing power, high-performance computing power, and power-saving at the time of receiving the order.
More than 1,000 users, from start-up companies to general electric appliances manufacturers, have conducted AI research and development utilizing cutting-edge technology.
In response to the rapid increase in demand for AI research and development in industry, government, and academia, AIST has been working on a higher-performance computing system while following the basic configuration of “ABCI” in order to strengthen “ABCI”. We will newly introduce an “AI bridging green cloud platform” that combines a large-capacity storage system. In addition to the introduction results of the existing “ABCI”, we have a system consisting of 120 latest PC servers equipped with 8 GPUs and a large-capacity storage system of 11.2 petabytes, and our company and Fujitsu Laboratories Ltd. (Note 5). The proposal that brings together the technical capabilities of) was highly evaluated, and we received an order for this system.
Features of “AI Bridging Green Cloud Infrastructure”
Large-scale cloud infrastructure for AI applications consisting of 120 latest servers
Eight high-end GPUs “NVIDIA A100” and two CPUs per node (3rd generation Intel) We will provide 120 next-generation models of “FUJITSU Server PRIMERGY GX2570” equipped with Xeon (R) scalable processor: development code name Ice lake). As a result, the theoretical peak performance of half-precision floating-point arithmetic is expected to be 850 petaflops, and that of double-precision floating-point arithmetic is expected to be 56.7 petaflops, which is the sum of “ABCI” and “AI Bridging Cloud Infrastructure”. This is equivalent to the 2nd place in Japan and the 6th place in the world in the latest supercomputer ranking TOP500 list in June 2020.
A large-capacity storage system that realizes integrated operation of the existing “ABCI” and “AI bridging green cloud infrastructure”
By linking the 11.2 petabyte large-capacity storage system installed in the “AI Bridging Green Cloud Infrastructure” with the already operating “ABCI” storage system and enabling mutual access, “ABCI” and “AI Bridging” Integrated operation of “Green Cloud Platform” is possible.
Fujitsu will continue to play a central role in contributing to the advancement of AI research and development in Japan through its involvement with the “AI Bridging Green Cloud Infrastructure” project with the AIST, leveraging the knowledge acquired through the construction and operation of this system for new initiatives with a variety of stakeholders in the industry, government, and academia.
(1) GPUs (Graphics Processing Units). A computing device for performing processing specialized in image processing and screen drawing.
(2) Half-precision floating-point operations A type of floating-point number that can be expressed on a computer in 16-bit form.
(3) Petaflops PFLOPS (Peta Floating-point Operations Per Second). Peta is 1,000 trillion (10 to the 15th power). 1,000 trillion floating-point operations per second.
(4) Double-precision floating-point arithmetic A type of floating-point number that can be represented on a computer in 64-bit form.
In the future, we will contribute to the development of AI research and development in Japan by providing the “AI bridging green cloud infrastructure” to AIST, and will apply the knowledge gained from the construction and operation of this system to industry, government and academia.