Give to WPI

ECE Graduate Seminar Lecture

Science / Technology - Lecture/Discussion - WPI Only

Thursday, February 6, 2014
3:00 PM-4:00 PM

Atwater Kent Laboratories
AK 233

Scalable Accelerator-Centric Clusters for High Performance Computing

The need for cost-effective, high-performance computation has become a common theme across diverse areas of science and engineering and its delivery a core issue in CISE research. At the same time, computer architecture has undergone an epochal change with the emergence of many-core CPUs, computational accelerators, and data-center/cloud computing. With it the focus in high-performance application development has shifted from riding increases in operating frequencies to achieving computational efficiency with available resources. In parallel, there has been a move to reduce cost--in power consumption, in system administration, and in amortizing capital outlay--through design and through the consolidation of computing capability. Both of these needs, increased computational efficiency and reduced cost, are addressed with the integration of communication and computation into accelerator-centric clusters with direct and programmable interconnects.

We describe work in the design and implementation of accelerator-centric clusters with direct and programmable communication. These are characterized by tightly integrated communication and computation and novel mechanisms to enable application-specific communication and application-aware network configuration. As a case study we present work on strong scaling with Molecular Dynamics.

Martin Herbordt
Associate Professor, ECE Department
Boston University

Martin Herbordt is an associate professor in the Electrical and Computer Engineering Department at Boston University. He received his B.A. degree in Physics and Philosophy from the University of Pennsylvania and his Ph.D. degree in Computer Science from the University of Massachusetts.

Since joining Boston University in 2001, a primary focus of Martins work has been various aspects of high-performance computing using accelerators such as FPGAs and GPUs: applications, including bioinformatics, molecular docking, molecular dynamics, and computational electrodynamics; development environments; and architecture of accelerator-centric clusters, i.e., compute clusters where accelerators communicate with each other directly.
This work is being supported by grants from the National Institutes of Health, the National Science Foundation, the Army Research Lab, and various industrial partners. Other projects underway include methods for power and thermal aware application development (supported by the MGHPCC) and HPC in the Cloud (with the Massachusetts Open Cloud). Recently completed work has involved fault-tolerant computation in space (supported by the Naval Research Lab) and in mapping algorithms to FPGA-based clusters (supported by the MIT Lincoln Lab).
Martin is a Fellow of the Hariri Institute for Computing and Computational Science and affiliated with the Center for Computational Science.

Host: Professor Yehia Massoud

Suggested Audiences: College


Last Modified: January 31, 2014 at 2:46 PM

Powered by the Social Web - Bringing people together through Events, Places, & Common Interests

  • Email a Friend
  • Bookmark this Page
  • Share this Page