Performance benchmarks?

Quantics development
1

Hello,

I am trying to look at the performance of the existing OpenMP in
Quantics, and I am looking for some suitable benchmarks.

I’m looking for a “standard” time evolution with no direct dynamics/
database interactions etc. A suitable benchmark might be something
that is a large enough problem size that it is (depressingly) slow
to make progress in the existing picture. The number of time steps
can usually be adjusted to provide a manageable wall clock time. I’d
need the relevant input files, and any insight on where you think the
problem may be would be useful (e.g., if you have a profile).

So, if you have something which you think might be interesting,
or if there are any questions, please let me know.

1 Like
2

Hi Kevin,

Here is a big calculation (thiophene photoexcitation) that could provide a good test. I haven’t run the full thing so do not know how long to expect it to take, but calculating the first energy on my PC took 2 mins and ca 4GB of memory. It is a straightforward MCTDH calculation and the profiling (in the timing file) shows that the calculation of the mean-fields dominates, as it should.

The .inp and .op files are attached as inp.txt and op.txt.

Let me know if that is a suitable place to start.

Graham

thio10d_inp.txt (2.3 KB)

thio_opt5_op.txt (70.0 KB)