Course in High Performance Computing TMA881, INN830

Larger and more complex mathematical models demand greater computer performance. There is a limit for what can be accomplished by one processor so to get really high performance or to solve very large problems a supercomputer must be used. This is not sufficient however. One must also know what algorithms are suitable for a special computer architecture and how to implement the codes, using system software, in an efficient way.

The main themes of this course are:

• Computer architecture. What are the building blocks of  a typical supercomputer?
   

• Algorithms fetched from the basic numerical analysis course. How come my Cholesky factorization runs at 10% of the theoretical top speed of my workstation? Are there faster codes?
   

• Programming workstations and parallel computers. So what are OpenMP, Lapack, MPI, PVM...?

The course contains several small assignments where you will test different algorithms, program packages, tools, languages and computers. Since the participants in the course usually have quite different backgrounds, we will start by looking at some basic tools such as make, ar, ld, prof and continue with a brief introduction to C and Fortran. This introduction will not make you a specialist in any of the subjects but it should be enough to take you through the assignments.

Most modern parallel computers consist of standard, off-the-shelf CPUs, connected by a fast network. So in order to get good performance on a parallel computer it is essential to understand how to tune codes for a uniprocessor

When it comes to parallel computers the course will cover message passing using MPI and shared memory computing (loop parallelism) using OpenMP.

This is a course for beginners in high performance computing and the prerequisites are basic courses in numerical analysis and programming, and perhaps the most important, you must think that computers are fun.

The literature will consist of lecture notes, manuals and articles.

Here is the formal course description (kursplan).

Thomas Ericsson, mathematics / numerical analysis
Tel: 772 1091
E-mail:  thomasmath.chalmers.se