[mpich-discuss] quad core mpich2 sluggish

[zach]

I installed the latest mpich2 with hydra and also configured the device to
sockets with no change.
I still get no performance improvement in speed in a four-process mpich run
over a single process run.

I tried some experiments:
Running without mpi i immediately see a drop in speed slowing by 80% when
starting a second independent run
I get the same speed regardless if i run 4 copies without mpi all at once or
submit a n=4 job with mpi with my computation grid 4 times wider
running one job without mpi takes as long as the same computational grid
with mpi with n = 4
its like it is not doing true parallelization even though i can see all cpus
utilized during the mpi run.

Someone else suspected I have some kind of memory bandwidth problem.
I am using the Q6600 2.4 ghz processor with an Asus p5n-e sli motherboard. i
have 8 gb ram
Could someone please help me understand if there is something about my
hardware that is indeed limiting?
I purchased this hardware hoping I could realize the x3-4 parallelization.
If it is my hardware what specifically do I need to look for in a future
purchase?
Is anybody else running mpich2 application on a quad core chip without
issues? If so what is the chip/motherboard?

Zach


----------------------------------------------------------------------
>
>
> > > I am running the latest ubuntu with a Q6600 quad core cpu and the
> mpich2 that came with ubuntu
> > > I can run my parallelized code over the four cores (i see it in 'top')
> > > however, the speed increase over a single process is hardly improved
> for 2,3, or even 4 processes.
> > >
> > > i was able to run the same code on a linux cluster through my school
> with speed increases as expected.
> > > am i doing something wrong?
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20110102/57356cd3/attachment.htm>