Benchmarking: Intel Linpack

Prerequisites

Make sure Hyperthreading is enabled in BIOS and this is run on Centos. Normally I have one SSD for windows and one for the Linux session saves time.

Head to Intel`s site to get the latest linpack libraries:

https://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download

Download the latest tgz for linux. This can potentially be done using wget (below is just a example, a newer version maybe available):

wget http://registrationcenter.intel.com/irc_nas/4547/l_lpk_p_11.2.0.003.tgz

How to run

Then untar:

tar -xvf l_lpk_p_11.2.0.003.tgz

The above should create a new folder of the untarred files, navigate into this and find main linpack x64 program:

cd linpack_11.2.3/benchmarks/linpack/

./xlinpack_xeon64

Then you should see:

Input data or print help? Type [data]/help :

Number of equations to solve (problem size):(here you put in your calculated problem size)
Leading dimension of array:(here you put in your calculated problem size)
Number of trails to run: (input the number of runs you want 3 for example)
Data allignment value (in Kbytes): 128

Linpack should then run after this.

To check that linpack is stressing the CPU`s and memory correctly. Open another session in Centos using keyboard shortcut "alt+f2" you should be presented with another login and password screen.

Login and type "top" this will show the process manager. If Linpack is stressing the system as expected then you will see CPU plus Memory usage rise up to match the problem size.

Depending on system Specification will vary the problem size required to make this as high as possible. Normally this is something between 80-100% CPU and memory usage.

Why not 100%? If you use a problem size that results in 90-100% usage I found the system can become unstable or the OS will just kill the process to maintain system stability. This is down to the OS requiring CPU time and ram of its own.