CS 758 Advanced Topics in Computer Architecture Fall 2019 Section 1 Meeting Time & Place: TuTh 930-1045 AM, 3444 Engineering Hall Instructor Matthew D. Sinclair URL: http://www.cs.wisc.edu/~sinclair/courses/cs758/fall2019/

Overview

Heterogeneous systems with programmable hardware accelerators seek to fulfill the promise of continued performance and energy-efficiency gains in the era of a slowing Moore's Law, larger problem sizes and an increased focused on energy-efficiency. These factors have caused hardware acceleration to become ubiquitous in today's computing world and critically important in computing's future. This course will explore some of the current research and industrial landscape of programmable hardware accelerators. This course qualitatively and quantitatively examines computer design tradeoffs for the architectures of programmable accelerators. We will learn about the architectures for these systems and study, for example, why different applications need very different accelerators to improve performance and energy efficiency. We will delve deeply into the architectures and modern models of modern massively parallel accelerators like GPUs and ML accelerators, culminating in a course project. Additional general topics in hardware acceleration will also be covered, including but not limited to augmented and virtual reality, graph analytics, genomics, and quantum computing. Finally, the course will also study the limits of accelerators and what level of fidelity is appropriate for modeling them, in order to identify the implications of future architecture designs.

Prerequisites

Examining tradeoffs requires that you already are well-versed in how a basic computer works, assembly language programming, pipelining, caching, and virtual memory. Thus, students are strongly encouraged to have this knowledge from prior courses (e.g., CS/ECE 552). A prior understanding of advanced computer architecture concepts such as out-of-order processing, branch prediction, cache coherence, and memory consistency (e.g., CS/ECE 752 and CS/ECE 757) will likely make it easier to understand the concepts in this course, but can be waived with instructor approval or taken concurrently. Similarly, advanced courses on parallel programming, including GPUs (e.g., CS/ECE/ME/EMA 759) may also be useful, although it is not necessary to take this course beforehand.