IRVINE, California – Dec 17, 2025 – Fixstars Corporation (TSE Prime: 3687, US Headquarters: Irvine, CA), NTTPC Communications, Inc. (NTTPC), and Getworks Co., Ltd. (Getworks) jointly conducted a Proof of Concept (PoC) to effectively manage the growing demand for high-performance GPU servers. The joint PoC achieved a high power usage effectiveness (pPUE*1) of 1.114, a level typically challenging to attain outside of large-scale data centers. This result confirms the commercial viability of utilizing liquid-cooled GPU servers within containerized data centers.
Through this verification, the companies successfully confirmed the commercial operation of a liquid-cooled GPU server within a dedicated, containerized data center. The PoC comprehensively demonstrated stable GPU operation, performance optimization, and improved power efficiency, resulting in enhanced operational efficiency.
*1 pPUE (partial Power Usage Effectiveness) is an indicator of power efficiency for a part of a data center (such as a module or a zone). It represents the ratio of the total power consumption of that specific area to the power consumption of the ICT equipment within it. While PUE measures the efficiency of the entire facility, pPUE indicates the efficiency of a partial scope. For modern, large-scale data centers using water cooling, the overall PUE typically ranges from 1.05 to 1.2, and a pPUE value around 1.1 is considered excellent. In this verification, pPUE is calculated as: (Power consumption including air conditioning and lighting in the containerized server room) / (Power consumption of all ICT equipment in the containerized server room).
Background and Challenges
The escalating adoption of generative AI and High-Performance Computing (HPC) is driving a rapid surge in the demand for high-performance GPU servers. Operating these powerful servers necessitates support for liquid-cooled GPU technology. While the commercial deployment of liquid-cooled GPU servers is expanding internationally, air-cooled data centers still predominate in Japan, leading to fewer domestic examples of liquid-cooled GPU server installations.
Moreover, achieving maximum GPU performance necessitates integrated collaboration across the data center, GPU server, and software layers. However, current optimization efforts frequently remain siloed within each individual layer.
To address these challenges and fully unleash the inherent capabilities of liquid-cooled GPU servers, the companies jointly initiated a Proof of Concept (PoC) utilizing a containerized data center.
Details of the PoC (Proof of Concept)
The PoC compared the performance of liquid-cooled and air-cooled GPU servers operating within a containerized data center environment. This involved applying various workloads and measuring a range of operational and performance metrics in an environment where the data center, GPU server, and software were fully integrated and optimized.
Verification Results
The PoC successfully validated the commercial viability of employing liquid-cooled GPU servers within a containerized data center. Furthermore, the PoC demonstrated the maximization of operational efficiency through integrated collaboration across the data center, GPU server, and software layers, achieving an outstanding pPUE of 1.114 for a liquid-cooled GPU server.
This verification successfully maximized the inherent capabilities of the liquid-cooled GPU server by leveraging a containerized data center, which enables the swift and flexible provision of the operating environment.
- A comparison of the liquid-cooled and air-cooled GPU servers in a containerized data center environment demonstrated an approximate 15-degree Celsius reduction in average GPU temperature under maximum load for the liquid-cooled system, confirming the proper function of the CDU and other components.
- A system was established to comprehensively collect and manage partial Power Usage Effectiveness (pPUE) metrics, which indicate the power efficiency of specific modules like the CDU and the data center room. This was achieved by real-time monitoring of server load and temperature, control of the InRow air conditioning*2, and implementation of containment*3 within the container.
- The measurement under the above environment resulted in a pPUE value of 1.114.
※2 In-Row Cooling: A localized cooling system installed between rows
of server racks.
※3 Containment: A partition or structural design used to efficiently separate cold air and exhaust heat to
improve cooling efficiency and maximize cooling performance in a data center.
* The full verification report for this PoC can be accessed via the following links.
- Thorough Verification of Liquid-Cooled GPU Server Effectiveness [Part 1]
https://www.nttpc.co.jp/gpu/article/benchmark27.html
- Thorough Verification of Liquid-Cooled GPU Server Effectiveness [Part 2]
https://www.nttpc.co.jp/gpu/article/benchmark28.html
Roles of Each Company
NTTPC
- Conducted this verification from the perspective of hardware engineering, leveraging their experience in providing multiple large-scale GPU clusters.
GetWorks
- Provided one 20-foot container-type data center, equivalent to a commercial environment, at the Yuzawa GX Data Center in Echigo-Yuzawa, Niigata Prefecture.
- Optimized the server operating environment through measures such as airflow containment during the verification.
Fixstars
- Provided a liquid-cooled GPU server and cooling equipment for the demonstration experiment. Built an integrated monitoring tool for comprehensive management of the environment, hardware, and software within the containerized data center.
Future Developments
As the adoption of high-performance GPU servers accelerates, the demand for liquid-cooled GPU servers is expected to expand even further in the future. Moving forward, the companies aim to further strengthen their collaboration and strive for the full-scale commercial deployment of liquid-cooled GPU servers in containerized data centers within Japan.
About the PoC
- Verification Period: Monday, October 6, 2025 – Friday, November 14, 2025
- Verification Location: Containerized Data Center equivalent to a commercial environment Getworks 20-foot containerized data center (Echigo-Yuzawa)
- Verification Servers: Dell PowerEdge XE9640 (Liquid-cooled configuration) / Dell PowerEdge XE8640 (Air-cooled configuration)
- Cooling System: AHx10 CDU (Manufactured by CoolIT Systems)
- Software: Fixstars AIBooster (Platform for optimizing AI processing performance)
About Fixstars Corporation
Fixstars is a technology company dedicated to accelerating AI inference and training through advanced software optimization solutions. It supports innovation in healthcare, manufacturing, finance, mobility, and other industries. For more information, visit: https://www.fixstars.com/
