Actual gains or losses also depend upon the memory and cache usage.
If your code is heavily eating cache and memory bandwidth, you get severe
contention conflicts between the threads or processes, and HT will degrade
the overall performance.
No miracle !
Pierre.
On Tue, 12 Oct 2004, Robin Humble wrote:
> On Tue, Oct 12, 2004 at 12:18:14PM +0800, L WK wrote:
> >>for sure.
> >> mpirun h -c 2 program
> >>you might get ~20% faster if you're lucky.
> >Thanks. It's said that to make use of hyperthread tech., you should have
> >some threads running inside your process, isn'it? Anyway, I have check this
>
> you need more than 1 thread or more than 1 process, yes. Otherwise how
> are you going to make use of your "extra" cpu?
>
> >But find that it just got a little bit faster than that running on a
> >non-hyperthreading-architecture system
>
> "a little" is about all you can expect. HT is just faking a 2nd cpu
> after all. you will never get 2x faster with HT. HT means you get
> slightly better usage out of your existing single cpu, assuming that
> the functional units (adders, multipliers, TLB etc) aren't maxed out
> already. For example, a super-efficient SSE/MMX code would see almost
> no speedup from HT.
>
> 10-20% speedup would be usual for "normal" code.
> This assumes that you don't introduce inefficiencies of the same order
> by breaking up your problem into 2 smaller parts (eg. crappy load
> balancing or becoming communication dominated).
>
> cheers,
> robin
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
--
Soutenez le mouvement SAUVONS LA RECHERCHE :
au national : http://recherche-en-danger.apinc.org/
sur Grenoble : http://recherchegrenoble.free.fr et http://etatsg.free.fr
_/_/_/_/ _/ _/ Dr. Pierre VALIRON
_/ _/ _/ _/ Laboratoire d'Astrophysique (UMR 5571 CNRS)
_/ _/ _/ _/ Observatoire de Grenoble / U. Joseph Fourier
_/_/_/_/ _/ _/ BP 53 F-38041 Grenoble Cedex 9 (France)
_/ _/ _/
_/ _/ _/ http://www-laog.obs.ujf-grenoble.fr
_/ _/ _/ mailto:Pierre.Valiron_at_[hidden]
_/ _/_/ Phone / Fax: +33 (0)4 76.51.47.87 / (0)4 76.44.88.21
|
