Homework:

You can turn in hard copy in class or e-mail the TA by midnight on the due date.

Homework 1 (pdf) (Due Monday, Februrary 23)

Homework 2 The Parallel Sorting Contest (pdf) (Due Monday, March 9)

More info on the Sorting Contest:

Sort two matrices of sizes 10,000 x 10,000*p and 100 x 100*p x 10,000, in the *p dimension. You can do additional examples if you have something cool implemented.
Use 8 processes on star-p. Since each node on evolution is a dual-core, you can optionally use 1 core per node (eight nodes) or four dual cores per node (four nodes).
To minimize your processes running over each other's, please use the following nodes according to your last names (x=0,1,2):
- A-I : tile-1-x, tile-2-x, tile-3-x
- J-P : tile-4-x, tile-5-x, tile-6-x
- S-Z : tile-7-x, tile-8-x, tile-9-x
How to control which nodes to use: Include only the nodes you want to use in the machine file in your ~/.starp/config directory on evolution (Details in a previous email). In the machine file, you can specify how many processes to assign to each host by adding ":n" after the node, where n is the number of processes. For example if you put "tile-3-0:1", it assigns 1 process to the node tile-3-0.

Please turn in a writeup by Friday, Mach 20

About evolution:

System has 30 nodes each with 2GB.
Each node has a single dual-core 2GHz Xeon processor (==> 8 GFlop/s peak per node).
Nodes have ethernet and Myrinet interconnect.

Also - GPU is NVidia 8800 GTX with 768MB on-card memory and peak flops ( 32-bit on 128 stream processors) of 518GFlop/s.

Project Proposal (Due Thursday, March 19)

Please turn in a short description of your course projects by Thursday night, or let us know if you are looking for suggestions.

See the Projects section for project ideas.