OpenMP and P4 hyperthreading

[Ole W. Saastad]

Will hyper-threading make sense when using OpenMP and more threads than
physical processors?

I have run the NPB2.3 benchmark in C/OpenMP version on a dual Pentium
Xeon
system and found some interesting results. For most of the benchmarks
there
is no gain in using hyperthreading, as expected, but the for the ep
benchmark 
there is a significant speed up. This benchmark contain a loop with a 
trancendentals like ln, exp and pow (pow is a combination of ln and
exp). 
The ep benchmark is supposed to scale almost perfect as it is
embarrassingly
parallel (hence the name ep), but it was somewhat unexpected that the
speedup
using four threads were so significantly.

For all the others there is a slowdown from 0 to 11%, but for the ep
there
is a speedup of 34%. 

The results can be viewed at :

http://computational-battery.org/

I have received a lot of comments about the hyperthreading due to my
former posting, but little actual benchmark results. It would be
interesting to see if there are other programs or problems that can
benefit from the hyperthreading. 

-- 
Ole W. Saastad, Dr.Scient. Scali AS P.O.Box 150 Oppsal 0619 Oslo NORWAY 
Tel:+47 22 62 89 68(dir) mailto:ole at scali.no http://www.scali.com 
Are you meeting Petaflop requirements with Gigaflops performance ?
    - Scali Terarack bringing Teraflops to the masses.