Preferred Networks, Inc. (PFN), Internet Initiative Japan Inc. ( IIJ ), and the Japan Advanced Institute of Science and Technology ( JAIST ) announced that they began trial operation of an ultra-efficient AI accelerator system in July at IIJ’s Matsue Data Center Park (Matsue DCP) and the JAIST Ishikawa Campus. This is part of the “Research and Development of Ultra-Efficient AI Computing Infrastructure” initiative, which was jointly proposed and accepted in the “Research and Development Project to Strengthen the Infrastructure for Post-5G Information and Communications Systems/Development of Post-5G Information and Communications Systems (Commissioned)” public offering by the Ministry of Economy, Trade and Industry and the New Energy and Industrial Technology Development Organization (NEDO). To achieve the high computing power required for its “Next-Generation MN-Core Series” AI processors, PFN has adopted a direct water-cooling system, a highly efficient thermal management system, for the first time, replacing the conventional air-cooling method. This system can efficiently handle the increased heat generated during operation as the processor’s computing power increases.
PFN developed a high-density server with a direct water-cooling mechanism that can accommodate eight MN-Core 2 boards. This testbed (test environment) began trial operation in July, with 30 nodes (240 MN-Core 2 boards in total) at the Matsue DCP and two nodes (16 MN-Core 2 boards in total) at the JAIST Ishikawa Campus. During this trial, IIJ and PFN will jointly install a high-density AI server with direct water-cooling in an air-cooled data center to verify solutions to the challenges inherent in water cooling. JAIST and PFN will also jointly conduct research on improving data center efficiency through the coordinated operation of water-cooling equipment and water-cooled servers (Software Defined Liquid Cooling Facility). In addition, PFN, IIJ, and JAIST will jointly conduct a demonstration experiment to verify the water-cooling technology and high-density performance of PFN’s next-generation MN-Core series.
IIJ is currently developing a modular data center compatible with direct water-cooling systems. As part of this demonstration experiment, the Matsue DCP’s air-cooled server room was equipped with a cooling technology called Air Assisted Liquid Cooling (AALC), which can be rapidly implemented using existing air-cooling equipment. This prototype hybrid cooling environment combines air-cooling and water-cooling technologies. IIJ plans to leverage the results of this demonstration experiment and development project to provide solutions to support the construction of AI infrastructure. During the trial operation, PFN will verify the direct water-cooling server for the next-generation MN-Core series, verify its efficiency and operability as a component of an AI computing infrastructure, and consider next-generation technologies. IIJ and PFN will verify solutions to the challenges inherent in water-cooling and verify interoperability between modules.
Also Read: Microsoft Japan Backs Growth of AI-Driven Frontier Firms
JAIST and PFN will conduct experiments to improve the efficiency of data centers through the coordination of water-cooling equipment and water-cooled servers (Software Defined Liquid Cooling Facility). Starting in fiscal year 2026, PFN, IIJ, and JAIST will conduct research and development on optimizing and streamlining resource allocation through the coordinated control of AI computing infrastructure using actual AI workloads in a testbed, with the aim of realizing an ultra-efficient AI computing infrastructure.
SOURCE: Yahoo