Benchmark page on wiki

You successfully prepared your system. Now: How do I tell Hippo to produce a trajectory from it? What parameters to choose? And what should I do when things go wrong.
Post Reply
oliver
Site Admin
Posts: 18
Joined: Fri Jun 27, 2008 8:18 pm
Location: Oxford, UK
Contact:

Benchmark page on wiki

Post by oliver » Fri Nov 14, 2008 2:53 pm

I started a simple Hippo Benchmark page on the wiki.

Basically, just run under Linux

Code: Select all

cd testjobs
time ./calc_testjobs
and report the user time and the cpu.

(Completely un-scientific but probably better than nothing and gives a first indication of relative performance of architectures.)

oliver
Site Admin
Posts: 18
Joined: Fri Jun 27, 2008 8:18 pm
Location: Oxford, UK
Contact:

Re: Benchmark page on wiki

Post by oliver » Mon Dec 08, 2008 3:22 pm

I added scaling benchmarks to the wiki page.

I simply ran the WALP + octane MD simulation, took the time and plotted the results. The general results (for this system) appear to be that performance gains beyond 4 cores are negligible.

oliver
Site Admin
Posts: 18
Joined: Fri Jun 27, 2008 8:18 pm
Location: Oxford, UK
Contact:

Conclusion from Benchmark page on wiki

Post by oliver » Mon Dec 08, 2008 3:26 pm

Tentative conclusions:
  • It does not matter if you use the hippo or hippo_p3 binary (at least for the revisions tested).
  • The MD code only scales up to ~4 cores (with a factor of about 3.3) (tested on the WALP+octane MD test case).
  • Fast Intel processors are good.
Last edited by oliver on Mon Dec 08, 2008 3:26 pm, edited 1 time in total.
Reason: fmt

Post Reply