next up previous
Next: Why Supercomputers?

HIGH PERFORMANCE COMPUTING I

E. Bruce Pitman

Lectures are Tuesday/Thursday 2-3:20.

HOT FLASH: An outline of homework and lectures over the last month can be found here

See

  • Patra on architectures
  • Patra on scaling

    Work expectations: Several homework sets will be assigned. Some of these will be pencil and paper 'problems', some will be coding assignments. These homework assignments will constitute 2/3 of the semester grade. There will be an end-of-semester project that will have both a code writing part and a write-up, and this project will count for 1/3 of the semester grade. There will be no final exam in this course.

    Outline: This semester we will concentrate on a few topics: (1) understanding (a bit) the design of modern high performance computers, including shared memory machines like the SGI Origin 2000 and distributed memory machines like the IBM SP and clusters, (2) programming for these architectures using OpenMP (on shared memory machines) and MPI (shared and distributed machines), (3) how some libraries (like BLAS) make efficient use of hardware to provide a framework for writing codes.

    I assume you have good coding skills in either FORTRAN or C (no C++). This is NOT a course in which to learn coding. If you are not comfortable writing code, please do not stay in this course - you will not enjoy it. A more appropriate course might be Mth 537 or Computational Physics. You do not need a background in differential equations for this first semester; in HPCII, familiarity with applications (and that usually means PDEs) is assumed. You do need some familiarity with numerical analysis/numerical methods.

    See COR501 from Fall 2000 and COR502 notes (on my courses page) and references therein.