Intel® Math Kernel Library 11.3 Update 4 Developer Guide

Ease-of-use Command-line Parameters

The Intel Optimized MP LINPACK Benchmark supports command-line parameters for HPL that help you to avoid making small changes in the HPL.dat input files every time you do a new run.

Placeholders in this command line illustrate these parameters:

./xhpl -n <problem size> -m <memory size in Mbytes> -b <block size> -p <grid row dimn> -q <grid column dimn>

You can also use command-line parameters to run prebuilt binaries and runme scripts. For example:

./xhpl_intel64 -n <problem size> -m <memory size in Mbytes> -b <block size> -p <grid row dimn> -q <grid column dimn>

./runme_intel64 -n <problem size> -m <memory size in Mbytes> -b <block size> -p <grid row dimn> -q <grid column dimn>

For more command-line parameters, see Heterogeneous Intel Optimized MP LINPACK Benchmark.

If you want to run for N=10000 on a 1x3 grid, execute this command, provided that the other parameters in HPL.dat are correct:

./xhpl -n 10000 -p 1 -q 3

By using the m parameter you can scale by the memory size instead of the problem size. The m parameter only refers to the size of the matrix storage and not to the coprocessor memory size or other buffers. So if you want to use matrices that fit in 50000 Mbytes with NB=1024 on 16 nodes, 32 nodes, and 128 nodes, execute these commands:

./xhpl -m 50000 -b 1024 -p 4 -q 4

./xhpl -m 50000 -b 1024 -p 4 -q 8

./xhpl -m 50000 -b 1024 -p 8 -q 16