REI101F High Performance Computing A
(Forritun ofurtölva A)

Fall 2012


News

Exam material available [Nov. 28th 2012]
New developments in the TOP500 list [Nov. 13th 2012]
A plan for the presentations has been announced below [Nov. 8th 2012]

General information

This course is the first in a series of two courses on high performance computing. The teacher for this first part is Hjálmtýr Hafsteinsson dósent. The purpose of this course is to introduce MS students in science and engineering to high performance computing machinery and get them used to writing parallel programs. The second course, REI102F High Performance Computing B, will not be held this winter. That course is more applied and usually has the students do parallel computing projects in their research areas.

There will be no fixed textbook for the course, but we will point you to useful books, and material on the web. See below.

Instruction in the class will be in english, unless all the students are Icelandic. However we will maintain this webpage in english and the homework will be in english. Icelandic students can do their homework in icelandic.

There will be weekly homework and two larger projects. The larger projects can be done in groups of two. The second project will be followed by a presentation of the results by the students. The homeworks will account for 15% of the final grade. Each project will count 15% and the presentation 5%. Alltogether 50%. Thus the final exam will be worth 50% of the final grade.

The class meets on mondays at 15:00 in classroom V-137 and wednesdays at 13:20 in classroom Á-423 according to the latest timetables.


Homework


Presentation schedule

Let me know if this list is incorrect or if you want to change dates.

Slides (only contain figures)


Interesting material related to the course


Benchmarks


Processor architecture

MPI material

OpenMP material

Jotunn fix

Jotunn can now be used without the queueing system, by issuing the command
        mpirun -mca btl ^openib -np 16 -npernode 4 -hostfile nodes mpitest
    
where nodes is a file listing all the nodes on Jotunn, and mpitest is an MPI executable. Also note that "-np" inicates the number of processes asked for, and "-npernode" is the number of processes that should be started on each node. The nodes on Jotunn have two cores and each of them is hyperthreaded, so they can easily handle 4 processes per node.

Useful books


hh (hja) hi.is, December 2012.