Shoaib Bin Altaf PhD Final Oral Defense

Thursday, December 8, 2016 - 11:00am
4310 CS

Speaker Name: 

Shoaib Bin Altaf

Speaker Institution: 

University of Wisconsin - Madison

Cookies: 

No

Description: 

Dissertation Title: Exploiting Simple Analytical Models for Modeling Hardware Accelerators

Committee:
David Wood, Advisor
Mark Hill
Mikko Lipasti
Karthikeyan Sankaralingam
Michael Swift

Abstract:

With the end of Dennard scaling, architects have increasingly turned to special-purpose hardware accelerators to improve the performance and energy efficiency for some applications. Unfortunately, accelerators don't always live up to their expectations and may under-perform in some situations. Understanding the factors which effect the performance of an accelerator is crucial for both architects and programmers early in the design stage. Detailed models can be highly accurate, but often require low-level details which are not available until late in the design cycle. In contrast, simple analytical models can provide useful insights by abstracting away low-level system details.

In this dissertation, we aim to explore the potential of such simple models for hardware accelerators. To this end, we propose two complementary proposals. In our first proposal, we develop a simple analytical model, LogCA, to argue whether an accelerator is helpful for
a given task or not. Once the usefulness of an accelerator is established, our second proposal, Accelerometer, helps in identifying bounds and bottlenecks associated with an accelerator design.

We validate our modeling framework across kernels of varying complexity on both on-chip and off-chip accelerators. We also describe the utility of our models using two retrospective case studies. First, we discuss the evolution of interface design in SUN/Oracle's encryption
accelerators. Second, we discuss the evolution of memory interface design in three different GPU architectures. In both cases, we show that the adopted design optimizations for these machines are similar to the suggested optimizations. We argue that architects and programmers can use insights from these retrospective studies for improving future designs.