|
Here are some sample test data for the matrix multiply project.
There are big and small data sets for big-endian processors such as
the IBM PowerPC (the Bluegrit cluster) and little-endian processors
such as Intel and AMD. The small data sets are only intended for initial testing as they are really too small for any kind of proper performance testing on the cluster. The big data sets take a significant amount of memory and about a minute to run on a single CPU, so any speedup on the cluster will be obvious. There is a MatLab script, MakeData.m, that you can edit and run to make datasets of whatever size you want. To use it, edit the array sizes in the script, run Matlab in a directory with MakeData.m, and type "MakeData". The Matlab script will print the time it took for the forward calculation. You may find it useful to generate test sets that are somewhere in between the two sizes I have provided. Note that each dataset uses the same filenames: Wset.dat, X.dat, and Y.dat. To keep track of multiple data sets, you may want to keep them in separate directories, rename them, use symlinks, or whatever. Here are links to the datasets: NOTE: You can download any of the scripts but only the small datasets can be downloaded here. Copies of the large datasets can be found on the IBM cluster in
/home/motteler/parpro/matmul/dataBE1 (Small IBM dataset)
/home/motteler/parpro/matmul/dataBE2 (Large IBM dataset
/home/motteler/parpro/matmul/dataLE1 (Small Intel/AMD dataset)
/home/motteler/parpro/matmul/dataLE2 (Large Intel/AMD dataset)
|