Towards Heterogeneous Systems for Edge Computing
Modern AI workloads combine computational steps with vastly different requirements as throughput, latency, and energyefficiency, making them a poor fit for homogeneous edge devices. The result is inefficient systems with high power consumption and limited capability. Heterogeneous architectures offer a solution: intelligently distributing workloads across specialized processors to match each step to the most suitable hardware. This talk explores the possibilities, design principles and practical challenges of building heterogeneous edge systems that deliver both computational power and energy efficiency.
- Lightning talk
Speaker
Giuseppe Sorrentino
PhD Candidate at Politecnico di Milano
Giuseppe is a PhD Candidate in Information Technology at Politecnico di Milano. His research focuses on accelerating compute-intensive workloads on heterogeneous systems by effectively combining multiple specialized hardware layers.... read more