Here’s the translation of the text into American English:
The new technology multiplies computational efficiency by up to 2.25 times in pilot tests and optimizes GPU usage for artificial intelligence applications.
Fujitsu has announced the launch of an innovative middleware technology called AI computing broker, designed to enhance the computational efficiency of GPUs in artificial intelligence (AI) applications and address the global shortage of these essential resources for AI processing. This middleware utilizes an adaptive GPU allocation system that optimizes resource usage and memory management in real-time, allowing multiple AI processes to run more efficiently.
After initial tests with companies such as TRADOM Inc. and SAKURA Internet Inc., the results showed an increase of up to 2.25 times in computational efficiency across various AI processes, enabling the handling of large volumes of data without requiring additional physical GPU capacity.
Resource Optimization for a More Efficient AI Future
Fujitsu’s technology allows for the simultaneous management of up to 150 GB of AI processing, employing advanced memory optimization techniques that quintuple the physical capacity of a typical GPU. The system dynamically allocates resources by GPU rather than by task, as is common in other systems, thus achieving greater availability and optimization of resources across different platforms and cloud environments.
This advancement is crucial in a context where demand for GPUs continues to rise due to widespread adoption of AI applications. According to the Japan Electronics and Information Technology Industries Association, the global market for generative AI is expected to grow 20 times between 2023 and 2030. However, meeting this demand presents a complex challenge that threatens to increase energy consumption in data centers, which are projected to account for 10% of global electricity consumption by 2030.
Collaborations and Successful Pilot Trials
To demonstrate the effectiveness of the middleware, Fujitsu has conducted trials in collaboration with several companies in the tech and finance sectors. These include:
- TRADOM Inc.: The Japanese company dedicated to risk management in the foreign exchange market will implement this technology to streamline the creation of AI models, achieving more accurate and faster results. Junichi Kayamoto, Director of Data Science at TRADOM, emphasized that “Fujitsu’s computing broker enables efficient multiplexing of AI processes, accelerating the creation of high-precision models.”
- SAKURA Internet Inc.: The cloud services company has begun a feasibility study of the system in its data centers to improve GPU availability. According to Ken Washikita, Director of SAKURA Internet Research Center, “this technology significantly enhances the efficiency of GPU allocation, allowing us to offer broader access to these resources.”
- AWL, Inc.: Specializing in AI camera solutions, AWL reported a significant reduction in GPU costs by applying Fujitsu’s middleware during the training of models in parallel, a crucial aspect for the company in its retail environments.
- Xtreme-D Inc.: The cloud services provider Raplase believes that the middleware will help its clients maximize the price/performance ratio in AI and HPC environments by improving GPU utilization in their data centers.
- Morgenrot Inc.: This company, a pioneer in container data centers, tested the technology with its cloud computing solution, achieving a 10% reduction in execution time by sharing GPUs among multiple tasks, enabling long training sessions and inference testing to occur simultaneously.
A Comprehensive Solution to a Global Problem
Fujitsu’s middleware addresses an urgent need to improve the efficiency and availability of GPUs, tackling the increased energy consumption in data centers and offering a solution that maximizes existing resources. Its application not only responds to the growing demand for AI but also helps reduce the environmental impact of data centers by optimizing energy consumption.
The technology will be available in Japan starting October 22, and its global rollout will occur progressively. With it, Fujitsu aims for AI service providers and companies operating in the cloud to maintain their competitiveness without compromising their resources or environmental impact, aligning with its sustainability commitment.
Fujitsu’s Commitment to Sustainability and Innovation
In its efforts to make the world more sustainable, Fujitsu continues to develop innovative technologies like the AI computing broker to optimize resource usage and reduce energy consumption in data centers. The Japanese company has established itself as a leader in the digital services market and is committed to the United Nations Sustainable Development Goals (SDGs), applying its technologies in key sectors to promote responsible and efficient digital development.
As Junichi Kayamoto from TRADOM stated, “this is a first step toward an AI that is not only powerful but also environmentally responsible.”
via: Fujitsu