Heterogenous Computing Systems

Evaluating the performance and energy-efficiency of heterogeneous computing systems

Presenter: Shubbhi Taneja

Presenter Status: Faculty

Department: Computer Science

Funding Source/Sponsor: RSCAP

Screenshot URL: https://drive.google.com/uc?id=17MJqXw18bTZeLInTrCBI2F5IhU8lVvA4

Abstract:
Accelerator-based systems that implement parallel computations have seen tremendous growth in the recent years for use in computation-heavy applications as they significantly reduce the computation time of High Performance Computing (HPC) applications. Graphics Processing Unit (GPU) based accelerators have become common for general-purpose computing beyond just 3D rendering. However, this comes at the cost of high power usage. Where CPUs are used for latency-critical applications, GPUs are better optimized for throughput-critical applications due to the large number of cores they have compared to CPUs. The goal of achieving higher performance by taking advantage of the strengths of both CPUs and GPUs simultaneously has created the paradigm of heterogeneous computing, which must be leveraged to meet the requirements of today’s computation-heavy applications