Colloquium -- Abhishek Bhattacharjee

Monday, April 2, 2018 - 11:00am
HFH 1132
Designing Efficient Heterogeneous Computer Systems Across Computing Scales
Tim Sherwood


Computer systems at all scales, from server-class systems for datacenters to

embedded systems on IoT devices, are embracing extreme heterogeneity in

hardware and software. While heterogeneity offers immense

computational promise, it also poses programmability and

performance/energy challenges. In this talk, I will show how we can leverage

decades of research on traditional general-purpose CPUs to improve the

programmability and efficiency of two classes of emerging heterogeneous

systems. In one example, we will improve the programmability and

performance of server-class GPGPUs using virtual memory techniques

developed over decades for traditional CPUs. In the second example, I will coopt

server-class hardware traditionally designed for branch prediction in

servers to instead manage energy in brain implants with a completely different

power/performance profile. At a high-level, these two examples represent two

types of heterogeneity -- intra- and inter-device -- and our work shows how we

can reap the benefits of specialization using modest hardware enhancements

of these systems.


Abhishek Bhattacharjee is an Associate Professor of Computer Science at

Rutgers, The State University of New Jersey. Abhishek’s research interests

are at the hardware/software interface, as it relates to the design of serverscale

systems for datacenters and embedded systems in IoT and biomedical

devices. He is the recipient of the CV Starr Fellowship from the Princeton

Neuroscience Institute, and the Rutgers Chancellor’s Award for Faculty

Excellence in Research.

Everyone welcome!