The widespread use of GPU for AI/DL workloads has furthermore increased the problem of effectively and efficiently allocating and using the GPU (possibly in a disaggregated context) while simultaneously achieving maximum efficiency of the usage of the resources and minimising the energy footprint of the workload.
A mechanism to minimise resource energy footprint for AI workloads by maximising the efficiency of the usage of the resource in disaggregated GPGPU clusters via advanced scheduling algorithms. The solution optimise the training's run-time Time to Solution of Deep Learning applications/AI workloads simultaneously with the system's energy consumption.
The solution enables companies and institutions running GPU-powered systems (e.g. service provdiers, cloud providers, data centres) to optimise the use of the systems with an easy to deploy, non-intrusive application. The solution is of interests for enterprises for deployment on their own systems, as well as for ISVs and Software Integrators to include in their own packages.