Fujitsu Unveils Breakthrough CPU/GPU Resource Optimization Technology
Fujitsu, a leading technology company, has introduced an innovative CPU/GPU resource optimization technology that aims to enhance the efficiency of high-performance computing (HPC) systems. This cutting-edge technology allows both CPUs and GPUs to work simultaneously, addressing the growing demand for GPUs in artificial intelligence (AI) workloads.
The resource optimization technology, claimed to be a world-first by Fujitsu, leverages the computational power of CPUs alongside the existing GPUs in a real-time switching mechanism. By automatically allocating the appropriate components for each task and seamlessly switching between them, this technology significantly improves the system’s ability to handle complex applications such as digital twins and generative AI.
Traditionally, program switching in HPC systems relies on unicast communication, where programs are executed sequentially on different servers. However, this approach suffers from variations in switching time, making real-time batch switching challenging. Fujitsu’s resource optimization technology solves this problem by utilizing broadcast communication, reducing the program switch interval from several seconds to just 100 milliseconds.
Let’s take an example of a system with one CPU and two GPUs running three programs. The technology initially assigns the first two programs to the GPUs. When the third program starts running, the system may redirect the first GPU to handle it instead. In response, the system assigns the first program to the CPU until the second program is completed. The GPU running the second program then switches its focus back to the first program, relieving the CPU of its workload. This dynamic allocation of resources ensures optimal performance and efficiency.
Fujitsu plans to incorporate this resource optimization technology into its future computer workload broker, a software initiative currently under development. This broker will enable AI systems to intelligently calculate and allocate resources based on specific customer requirements, including computation time, accuracy, and cost.
Furthermore, Fujitsu intends to implement this technology in its AI platform, known as Fujitsu Kozuchi, which heavily relies on GPUs for processing. Additionally, the company will utilize the resource optimization technology in its 40-qubit quantum computer simulator.
By introducing this breakthrough technology, Fujitsu aims to bridge the gap between the increasing demand for high-performance GPUs and their limited supply. The resource optimization technology not only enhances the capabilities of HPC systems but also paves the way for more advanced AI applications and quantum computing simulations.
In conclusion, Fujitsu’s CPU/GPU resource optimization technology represents a significant advancement in the field of high-performance computing. By harnessing the combined power of CPUs and GPUs and enabling seamless switching between them, this technology enhances system efficiency and opens up new possibilities for AI and quantum computing.
