Fujitsu introduces a new technology that can distribute computing time for CPU and GPU in real time depending on the needs of the applications. In theory, this means that all available capacity is used optimally.

Fujitsu is proposing a new technology with which it wants to optimally distribute the computing power of the CPU and GPU across applications. Fujitsu says its new tool is the first that can distribute computing time in real time according to the needs of applications. The Japanese company hopes that the solution will enable more efficient use of the existing hardware. The technology thus offers a perspective for the global GPU shortage.

Optimal distribution

CPUs and GPUs can serve different applications at the same time, but that does not mean that the capabilities of both components are optimally utilized by default. Fujitsu gives an example in which three applications compete for the computing time of two GPUs in a server.

In such a scenario, it is possible to assign GPU one to app one and GPU two to app two. When app three is launched, the software temporarily uses one of the GPUs to speed up the application’s workload. The results of this measurement are compared to previous measurements in apps one and two. It might then turn out that app three benefits more from GPU acceleration than app one, reducing the overall computing time on the server when GPU two is assigned to app three. Then when app two is ready on GPU two, app one can use the freed graphics card.

Real-time priority

In HPC systems, the software can also switch between different programs in real time, so that, for example, code that needs to be executed in real time is given priority. Fujitsu says this is useful for applications in digital twins, generative AI and more. In both cases, the situation is similar: software demand for GPU acceleration exceeds supply, so Fujitsu’s technology acts as an arbiter to distribute computing power as efficiently as possible.

Fujitsu initially wants to implement the new technology in its own solutions. The Japanese are initially looking at the Fujitsu Kozuchi AI platform, but the company’s HPC service would also benefit from it.