This research line tackles the design and management of power-efficient high-performance computing servers and data centers running compute and memory intensive next-generation workloads. These workloads include AI analytics, Deep Learning training and inference, Quality-of-Service constrained applications such as video transcoding, or next-generation genome sequencing.
We use machine-learning based resource management and task mapping techniques to run a certain workload in the most efficient way (in terms of performance per watt). These techniques are implemented with constant awareness of the underlying heterogeneous hardware resources.
Power-aware acceleration of Deep Learning (DL) training and inference on High-Performance Computing (HPC) servers
Resource management from the edge to the cloud and efficient simulation of Internet-of-Things (IoT) scenarios for Artificial Intelligence (AI) applications
Multi-objective machine-learning based resource management for heterogeneous HPC servers and data centers