Fujitsu, a global leader in technology and innovation, has announced a groundbreaking technological advancement that promises to revolutionize the utilization of CPUs and GPUs. This cutting-edge technology, the world’s first of its kind, dynamically allocates computing resources ...
Fujitsu, a global leader in technology and innovation, has announced a groundbreaking technological advancement that promises to revolutionize the utilization of CPUs and GPUs. This cutting-edge technology, the world’s first of its kind, dynamically allocates computing resources in real-time, prioritizing processes with high execution efficiency. This development comes at a critical juncture as the world grapples with a severe shortage of GPUs, exacerbated by the surging demand for generative AI, deep learning, and other GPU-intensive applications.
Fujitsu’s innovative solution promises to optimize existing computing resources and offers a lifeline to industries and researchers reliant on high-performance computing (HPC). The company has additionally pioneered a real-time parallel processing technology, enabling the seamless execution of multiple programs on HPC systems without waiting to complete ongoing processes. This breakthrough is poised to significantly enhance the efficiency of large-scale computations, particularly in fields such as digital twin modeling and generative AI applications.
Addressing the Global GPU shortage
The global shortage of GPUs has presented a major challenge to industries and researchers reliant on these processing units for intensive applications such as generative AI and deep learning. Fujitsu’s solution tackles this issue head-on by efficiently allocating CPU and GPU resources in real-time, ensuring that high-priority processes receive the computational power they need, even in a multi-program environment.
Utilizing CPU and GPU during program execution
Fujitsu’s technology distinguishes between programs that require GPU acceleration and those that can be processed by a CPU, all while multiple programs run simultaneously. It achieves this by predicting the rate of acceleration and allocating GPUs to high-priority program processing in real-time. Here’s a simplified example of how it works:
1. The user aims to process three programs using one CPU and two GPUs.
2. GPUs are initially assigned to programs 1 and 2 based on availability.
3. In response to program 3’s request, GPU allocation switches from program 1 to program 3 for performance measurement.
4. The system measures the degree of processing acceleration on the GPU, finding that allocating the GPU to program 3 would reduce overall processing time.
5. Consequently, the GPU is allocated to program 3 while the CPU is allocated to program 1. After program 2 finishes, the GPU is reassigned to program 1, optimizing computational resources and minimizing processing time.
This groundbreaking technology is particularly valuable for accelerating tasks like training AI models and processing graph AI data, ultimately expediting the development of AI applications and advanced image recognition.
Real-time switching of execution in HPC systems
In addition to addressing the GPU shortage, Fujitsu has developed a pioneering technology that enables real-time switching between multiple programs on HPC systems. Unlike traditional methods that wait for one program to conclude before switching, this innovation allows for immediate program execution without delay.
Conventionally, HPC systems employed unicast communication, which switched program execution sequentially, leading to timing variations and challenges in achieving real-time program execution. Fujitsu’s solution leverages broadcast communication, enabling simultaneous communication to switch program execution. This reduces the interval between program processing switches from several seconds to just 100 milliseconds in a 256-node HPC environment. Users can select the optimal communication method considering application requirements and network quality, thus ensuring the best performance.
This technology paves the way for rapidly executing applications demanding real-time performance, such as digital twin modeling, generative AI, and materials and drug discovery, using HPC-like computational resources.
Plans and implications
Fujitsu intends to incorporate the CPU/GPU resource optimization technology into its Fujitsu Kozuchi (code name) – AI Platform. This platform enables users to swiftly test advanced AI technologies, further accelerating AI research and development.
The HPC optimization technology will be applied to Fujitsu’s 40-qubit quantum computer simulator for collaborative computing, enhancing its capabilities and potential applications.
Fujitsu also has its sights set on broader applications for its technology. Fujitsu Computing as a Service HPC provides users access to cost-effective, high-performance computing resources for simulation, AI, and combinatorial optimization problems. Additionally, Fujitsu is exploring the application of this technology to the Composable Disaggregated Infrastructure (CDI) architecture, allowing users to adapt hardware configurations among servers easily. These endeavors align with Fujitsu’s mission to create a more accessible, sustainable, high-performance computing ecosystem.
About Fujitsu
Fujitsu’s mission is to make the world more sustainable by building trust in society through innovation. As the preferred digital transformation partner for customers across over 100 countries, Fujitsu’s 124,000 employees are dedicated to addressing some of humanity’s most pressing challenges. Fujitsu’s services and solutions draw from five core technologies: Computing, Networks, AI, Data and security, and Converging Technologies. By leveraging these technologies, Fujitsu aims to drive sustainability transformation.
Fujitsu Limited (TSE: 6702) reported consolidated revenues of 3.7 trillion yen (US$28 billion) for the fiscal year ending March 31, 2023, solidifying its position as Japan’s leading digital services company by market share.
Fujitsu’s groundbreaking technology offers hope amid the global GPU shortage, promising to optimize CPU and GPU resource allocation in real-time. With the ability to distinguish between CPU and GPU processing needs, this technology has the potential to revolutionize AI development and high-performance computing. Furthermore, Fujitsu’s real-time switching capability for HPC systems opens new horizons for applications requiring immediate program execution. As Fujitsu continues to explore wider applications for its innovations, it aims to foster a society where high-performance computing resources are accessible to all, driving innovation and sustainability into the future.