Software Archive
Read-only legacy content
17061 Diskussionen

Xeon Phi Stress Test

kbzowski
Einsteiger
1.885Aufrufe

I would like to check temperature during highest possible utilization rate of MIC card for long time. The goal is to check my custom made cooling system before moving workstation to production. I have already tried linpack from Intel Parallel Studio XE but utilization (using default settings) was not so high. In other words - tests were too fast to check my cooling solution.

What can you suggest to perform simple stress test of Xeon Phi?

0 Kudos
1 Lösung
Andrey_Vladimirov
Neuer Beitragender III
1.885Aufrufe

For thermal testing you can use the package "micperf", which is included with MPSS. To install micperf in MPSS 3.1, run from the directory where you untarred the MPSS tar ball:

[bash]

sudo yum install perf/micperf*.rpm

export PYTHONPATH=/usr/src/micperf/micp:${PYTHONPATH}

export PATH=/usr/src/micperf/micp/micp/scripts:${PATH}

[/bash]

To run a stress test on mic0 using DGEMM for approximately 1 hour:

[bash]

micprun -k dgemm -d 0 -p "--i_num_rep 3000"

[/bash]

To monitor the temperature, in another terminal window run:

[bash]

watch -n 1 micsmc -t

[/bash]

To run a stress test on multiple coprocessors, you will have to run several instances of "micprun" in parallel.

Lösung in ursprünglichem Beitrag anzeigen

5 Antworten
TaylorIoTKidd
Neuer Beitragender I
1.885Aufrufe

Have you disabled all power management?

Have you looked at the core utilization to insure that you are fully utilizing all the cores?

--
Taylor
 

Andrey_Vladimirov
Neuer Beitragender III
1.886Aufrufe

For thermal testing you can use the package "micperf", which is included with MPSS. To install micperf in MPSS 3.1, run from the directory where you untarred the MPSS tar ball:

[bash]

sudo yum install perf/micperf*.rpm

export PYTHONPATH=/usr/src/micperf/micp:${PYTHONPATH}

export PATH=/usr/src/micperf/micp/micp/scripts:${PATH}

[/bash]

To run a stress test on mic0 using DGEMM for approximately 1 hour:

[bash]

micprun -k dgemm -d 0 -p "--i_num_rep 3000"

[/bash]

To monitor the temperature, in another terminal window run:

[bash]

watch -n 1 micsmc -t

[/bash]

To run a stress test on multiple coprocessors, you will have to run several instances of "micprun" in parallel.

kbzowski
Einsteiger
1.885Aufrufe
Andrey - this was exactly what I was looking for. Thank you.
Pawel_E_
Einsteiger
1.885Aufrufe

with MPSS 3.8.5 what am I missing?

$ micprun -k dgemm -d 0 -p "--i_num_rep 3000"
WARNING: unable to determine MPSS version from installed RPM's, setting to 3.1
[ OFFLOAD     ] native
ERROR:  
    Make sure that compilervars.sh from Composer_XE is sourced in
    your working environment.  The Composer_XE_2017 redistributable
    package located here:
    https://software.intel.com/en-us/articles/redistributables-for-intel-parallel-studio-xe-2017-composer-edition-for-linux
    can be installed to access the required shared object libraries.

I have compiler suite verion compilers_and_libraries_2019.4.243 installed but sourcing compilervars.sh changes nothing.

Some suggestions?

Many thanks, L.

JJK
Neuer Beitragender III
1.885Aufrufe

for the Xeon Phi coprocessor , you need the 2017 libs, as Intel dropped support for the Xeon Phi (KNC) in later versions.

I've just grabbed the latest 2017 libs and the command you posted "just works".

Antworten