Hi Robin
As far as I know, the HyperThreading capable Intel processors have no
additonal floating poitnt execution
units, so if your application is numerically intensive, then (in theory)
there will be no performance improvements
even if the scheduler makes good work. When you are using a program
which makes integer calculations (and
hasn't got memory bottleneck eg. compiling), then you should gain some
benefit from HT.
My personal opinion is that widely used scientific codes doesn't really
work well with HT (it is not designed
for this purpose).
Probably the best thing you should do, is to give it a try and see what
happens. If you do so then you should use
2.6.x Linux kernel series, because of the better HT support and
advanced scheduling techniques.
Regards
Andras Szabo
Robin Humble wrote:
>On Thu, Jun 24, 2004 at 09:15:13PM -0400, Yaron Minsky wrote:
>
>
>>One important thing is if you're not going to put 4 processors per
>>node (which isn't worth it in my experience), then turn hyperthreading
>>off. If you don't, then depending on the scheduler you may get two
>>
>>
>
>definitely.
>If changing HT settings in BIOS on a whole cluster doesn't appeal, then
>with oldish 2.4 kernels (< 2.4.22?) the 'noht' boot option works, and
>with new 2.6 kernels 'maxcpus=2' is the boot option.
>for kernels in between, compiling with CONFIG_NR_CPUS=2 most likely
>will work (but you should test it).
>
>cheers,
>robin
>_______________________________________________
>This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
>
>
|
