ABINIT Discussion Forums

Greetings,
I'm a sysadmin and recently I've just finished compiling the latest version of abinit (6.8.1) on one of our compute clusters. One of the worker nodes have some nVidia Tesla GPUs and I compiled with the '--enable-gpu' flag, as described in the documentation.
Now I would like to run a test or two in order to verify if my abinit installation is correctly linked to the CUDA libraries. I've ran some existing jobs and the results seem valid, but there was no load on the GPU units. I assume this is because that particular job did not make use of the GPU enabled code in abinit.

I looked in the documentation and online but I am not able to find a test that I can run that will exercise the part of abinit that has be optimzed to use GPUs.

Does anyone have any recommendations or suggestions?

Thanks a lot!
Andre

GPU support is enabled for recursion. You can test it doing the following:


Test T in the 'paral' series also uses this part of Abinit.

Hello Andre,
the test proposed by Pouillon is good but to appreciate the GPU
I would suggest to make another test.
After to have executed the test described (but only one):
make tests_v5 start=75 stop=75
you should copy locally the t75.in file (look in the ab.files for its location),
change the value of the variable ngfft (for example: ngfft 3*24 => 3*36 => 3*48 =>3*64),
add the line: "timopt -1"
Finally change the first line of the ab.files (to look at the modified t75.in)
To launch the tests: abinit <ab.files > logtest

NB1- To have an idea of the speedup you should compile abinit also without gpu and make the same tests.

NB2- In the standard ouput file the only reference to GPUs appears in the timing otherwise the reference in GPU will be only in the log file.

Regards,
Marco