[hcs-d] TALK: Analysis-Driven Performance Optimization with CUDA (Cliff Woolley from NVIDIA, Tuesday April 5h, 8:35 PM, Harvard MD G125)

Nicolas Pinto pinto at mit.edu
Sun Apr 3 19:21:58 EDT 2011

Title:  Analysis-Driven Performance Optimization with CUDA
Speaker:  Cliff Woolley (CUDA Developer Technology Engineer, NVIDIA Corporation)

Date: 4-5-2011
Time: 8:35 PM
Location: Harvard Maxwell Dworkin G125 (http://j.mp/eCgV66)

Harvard CS264 2011 Guest Lecture Series
"Massively Parallel Computing" Course
Host: Nicolas Pinto (Harvard, MIT)


The goal of this talk is to demystify performance optimization by
transforming it into an analysis-driven process. There are three
fundamental limiters to kernel performance: instruction throughput,
memory throughput, and latency. In this talk we will describe how to
use profiling tools and source code instrumentation to assess the
significance of each limiter, what optimizations to apply for each
limiter, and how to determine when hardware limits are reached.

Notes:  It is assumed that attendees are already familiar with
fundamental optimization techniques.  This talk was originally
prepared and delivered by my colleague, Paulius Micikevicius, for

Speaker biography:

Cliff Woolley is a CUDA Developer Technology Engineer with NVIDIA
Corporation.  He received his Master's degree in Computer Science from
the University of Virginia in 2003.  He was among the earliest
academic researchers to investigate the use of graphics processors for
general purpose computation, having applied these early GPGPU ideas
both to non-traditional graphics rendering techniques as well as to
non-graphical algorithms such as a multigrid solver for PDEs.

Nicolas Pinto, PhD

More information about the hcs-discuss mailing list