yade-dev team mailing list archive

[Václav S(milauer <eudoxos@xxxxxxxx>]

> export OMP_NUM_THREADS=2; yade-trunk-opt simple-scene-parallel.py
> 36114.1206212 iterations/sec
>
> export OMP_NUM_THREADS=1; yade-trunk-opt simple-scene-parallel.py
> 112233.445567 iterations/sec
>
> yade-trunk-opt simple-scene.py
> 105485.232068 iterations/sec
>
>
>   
What is your hardware? I get to 5e4 iters/sec on my laptop, probably to
double on 2 x dualcore opterons (didn't try). (Sadly, Playstation 3 has
max. 256MB of memory, otherwise it would be killer thing to run
computations on - with 7 cores for massively vectorized floating point...)

openMP defaults to number of threads = number of cores (IIRC).... Second
and third cases being roughly equal would mean you run on 1 core
machine? No idea.

There is some overhead of starting the parallel section, which in the
case of simple-scene probably ouweights the gains of it. I suspect
however that it might be worth it for larger simulation, where we are at
few iterations / sec.

Vaclav
_______________________________________________
yade-dev mailing list
yade-dev@xxxxxxxxxxxxxxxx
https://lists.berlios.de/mailman/listinfo/yade-dev