Course Info
Lectures | Tuesday 10:15-12:00 in INR 219 & Thursday 15:15-17:00 in INF 213 |
---|---|
Web Page | http://parsa.epfl.ch/course-info/cs471/ |
Instructor | Babak Falsafi |
Email, URL | babak.falsafi /at/ epfl.ch, http://parsa.epfl.ch/~falsafi/ |
Office | INJ 233 |
Office Hours | By appointment |
Phone | +41 21 693 55 92 |
Assistant | Mark Sutherland |
Email, URL | mark.sutherland /at/ epfl.ch, https://sites.google.com/site/markjohnsutherland/home |
Office | INJ 239 |
Office Hours | Tuesday 15:00-17:00 |
Admin. Assistant | Stephanie Baillargues |
stephanie.baillargues /at/ epfl.ch | |
Office | INJ 234 |
Advanced Multiprocessor Architecture
We have witnessed four decades of Moore's law (scaling chips in density) accompanied with Dennard scaling -- i.e., a reduction of voltages that would allow for larger chips with minimal to no increase in total power. Unfortunately, Dennard scaling has slowed down (starting mid-2000's) forcing designers to divide chips up into multiple processing cores. Thus, multiprocessors became the defacto building blocks for all computers. Recently, the advent of machine learning has led to the adoption of accelerators. The result is the emergence of heterogeneous computing systems featuring multiprocessors and accelerators. This course will build upon the fundamental concepts offered in computer architecture and introduction to parallel architectures to cover modern design techniques as well as projections of designs into the next decade. We will cover topics as diverse as modern memory hierarchies and interconnects, GPUs, and machine learning accelerators.Topics addressed by this course include:
- Performance Metrics
- Programming Models
- Cache Coherence
- Memory Consistency
- Synchronization
- Transactional Memory
- CMP Caches
- Interconnection Networks
- Scaling Trends
- Server Chips
- Specialization
- GPU
- Machine Learning Accelerators
- Emerging Memory
- Distributed Memory
- Datacenters
Who should take CS 471?
CS 471 is a graduate course and is highly recommended for master and PhD students. Like other graduate-level courses, the course includes weekly readings, discussions, and questions on papers of seminal and recent contributions to the field of computer architecture. The course also includes a research project, in which students study, improve, and evaluate multiprocessor innovations. A list of project ideas will be given out, but students can suggest and work on their own ideas with potentials for advancing the state of the art. Feedback on performance will be given only upon request by a student. There will be no recitation classes.Attendance
You are responsible for all the material covered in class including handouts and class notes. If you are unable to appear for an exam for any reason, you must contact Prof. Falsafi before the exam. The course will be held in person, with a live Zoom session running for each lecture. The Zoom link for the course can be found here.Prerequisites
CS-208 Computer architecture / Architecture des ordinateurs, CS-206 Parallelism and concurrency, and CS-307 Introduction to multiprocessor architectureHomework & Paper Review
All assignments will be available on the course home page. We do not accept late homework. The homework will consist of questions on the weekly reader.Project
There will be a research project due at the end of the semester. The project handout will be distributed in the first week of the semester. You will have to find a project partner and conduct research in groups of two. The project will account for a substantial fraction of the overall grade.Grading
A tentative breakdown of the grades (subject to change) is given below.
Homework/Review Project Midterm Final Participation & Discussion |
25% 25% 25% 25% Counts |