About ABCI
AI Bridging Cloud Infrastructure (ABCI), is an open computing infrastructure for both developing AI technology and bridging AI technology into the industry and the real world, constructed by National Institute of Advanced Industrial Science and Technology (AIST), and operated by AIST Solutions Co. (AISol). ABCI started full-scale operation in August 2018, was upgraded to ABCI 2.0 in May 2021, and is scheduled to be upgraded to ABCI 3.0 in January 2025.
Largest in Japan, Super Energy Saving, Open AI Infrastructure
As part of the Ministry of Economy, Trade and Industry’s “Global Research Center Development Project on Artificial Intelligence” (H28 Secondary Amendment), National Institute of Advanced Industrial Science and Technology (AIST) has developed from scratch on AIST Kashiwa Center within the Kashiwa II Campus of the University of Tokyo. Since then, ABCI has been expading its computing resources with support from “Bridge Infrastructure Expansion on Artificial Intelligence” (R1 Amendment) and “Establishment of Computing Resources Contributing to Strengthening the Fundamental Development Capabilities of Generative AI” (R5 Amendment).
As an open innovation platform to accelerate AI research and development by industry, academia, and government in Japan, ABCI aims to research, develop, and demonstrate AI technologies, promote social implementation, and take on the most important challenges in the field of AI by utilizing one of the largest computational capacities in Japan.
Due to the need to foster fundamental research and development capabilities in Japan, particularly in the field of generative AI, ABCI has focused on allocating computational resources to accelerate the development of generative AI since fiscal year 2023. From fiscal year 2024 onwards, in order to strengthen research and development capabilities in cutting-edge AI technologies, including generative AI models, ABCI will provide computational resources primarily for public use by domestic industry, academia, and government, particularly national research institutes and universities.
Major Features of ABCI
Designs ultra-high-density, ultra-power-efficient data centers from scratch
A low-cost, lightweight “Warehouse” was built, and a double-layered structure was designed to assemble “Spires” inside it, achieving a thermal density several times greater than that of a normal data center. For cooling, we are trying to utilize a cooling system for supercomputers for consumer use. The combination of hot water cooling and air cooling enables 70 kW of cooling per rack while drastically reducing the power required for cooling. As a result, the system also helps to reduce operating costs.
Adoption of de facto standard architecture
We consistently develop and provide commodity hardware, especially AI accelerators, which are the de facto standard in AI development. We design our products to be easy for private cloud providers to emulate and cost-effective for users.
Leveraging the Software Ecosystem
We provide a software development and execution environment with state-of-the-art middleware, parallel compilers, the latest AI accelerator development environment, and deep learning tools. In addition, to enable efficient AI development by leveraging the software ecosystem, ABCI has been providing a container environment since the beginning of the ABCI service, which enables the use of a variety of software assets developed by the global community, including ABCI-optimized containers.
The ABCI can be accessed through a web browser. By using Open OnDemand, it is possible to perform console operations, file operations, and use web applications, such as Jupyter Lab, on the web browser.
The ABCI User Guide provides more information about the software offered by ABCI.
Promoting the Secure Use of Data
Since data is critical to the development and use of AI, ABCI provides multi-petabyte scale high-speed storage and data infrastructure for sharing and publishing data. Communication channels and data can also be encrypted using de facto standard cloud technologies. In addition to this, ABCI is promoting an experiment called the ABCI Data Set, which catalogs users’ data and learned models.
