Course Description

This course on Advanced Computer Architecture II covers parallelism and the design of parallel computers. Historically, parallel computers have been designed for the sole purpose of quickly solving large-scale computational problems like weather forecasting or molecular modeling (to name just two examples). These problems are usually expressed as a series of floating-point computations of large data sets stored in multidimensional arrays, and can usually be partitioned across multiple processors to achieve large-scale parallelism. However, within the last fifteen years, new applications for parallel computers have eclipsed these traditional numeric codes, and are the driving force behind the tremendous volume and revenue growth in the marketplace for parallel computers. These applications span all the way from commercial server workloads that run in managed datacenters, to heavily-threaded games and web browsers running on PCs and laptops, to massively data-parallel applications like graphics rendering. In other words, parallelism in applications and in hardware has become pervasive in our industry. This course will study the nature of parallelism across these application domains, with focus on the hardware required to support parallel execution. We will investigate techniques for detecting, increasing, and exploiting parallelism and will study in detail the design of various components of parallel computer systems. We will read research papers that study state-of-the-art designs and explore exciting new ideas in a course project.

More details can be found in the Course Syllabus (on Canvas).

Prerequisites:

ECE/CS 552 (or equivalent).

Note: ECE/CS 752 is not a prerequisite for this course.

Grading:

16% Paper Reviews

5% Quizzes

20% Midterm Exam 1

20% Midterm Exam 2

34% Project

5% Class Participation

There is no required textbook; we will rely on readings from the literature and other online resources. This book is a useful additional reference: "Parallel Computer Organization and Design" by M. Dubois, M. Annavaram, and P. Stenstrom."