We motivate high performance computing by showing that numerical experiments are becoming a third pillar of the scientific method, complementing theory (because many phenomena are too complicated to understand theoretically) and conventional experiments (because many phenomena or devices are too difficult, expensive, slow or dangerous to measure or prototype in the laboratory). Second, we motivate the need for parallelism, by showing that despite the exponentially increasing power of serial processors over time, using many of them in parallel is essential to solve very large problems. Some trends in high performance computing are discussed with data from the TOP500 list. Third, we described the challenges in writing and understanding parallel programming, a more difficult activity than conventional serial programming. Finally, we outlined the structure of the course.
PowerPoint
Sourcebook, Chapter 1
The Marketplace of High-Performance Computing, Erich Strohmaier, Jack J. Dongarra, Hans W. Meuer, and Horst D. Simon. Parallel Computing, Vol. 25 (13-14) (1999) pp. 1517-1544. pdfTake a look at the TOP500 list.
Sign up for a Millennium account (PI: Kathy Yelick (yelick@cs.berkeley.edu), Description of Research: CS 267).Sign up for a NERSC account (see the Jason).
Mail in your survey to complete the student information.
Assignment #0 is due 1/28/2004
1) Why is computation becoming the third pillar of science?
2) Give an example of a limitation of traditional experimental science that can be overcome be computational science.
3) What is Moore's Law? What are ultimate limitations of Moore's Law?
4) Why is parallelism used in computer architecture?
5) Give several examples of parallelism in computer architecture.
6) How do superocmputers of today look like?
7) What is the fundamental problem that we are trying to address in this class?
8) What is LINPACK performance?