| Re: Hyperthreading and IDL? [message #41142] |
Wed, 29 September 2004 12:51 |
Marc R. Reinig
Messages: 6
Registered: May 2004
|
Junior Member |
|
|
"Edd Edmondson" <eddedmondson@hotmail.com> wrote in message
news:cje5vn$m5o$2@news.ox.ac.uk...
> Jonathan Greenberg <greenberg@ucdavis.edu> wrote:
>> Has anyone done any performance tests looking at IDL 6.0 (or 6.1) under
>> hyperthreaded architectures? I see that certain processes max the
>> processor
>> out at exactly 50%, which makes me think that I'm getting a 50%
>> performance
>> hit just using the hyperthreading enabled. Thoughts? Experience?
>
> It's probably a figment of the way the hyperthreaded CPU appears to the
> kernel - as one CPU or as two or whatever. It'd be best to do some sort of
> quick benchmark. Use 'timer=SYSTIME(/SECONDS)' before some long routine
> and then do a 'print, SYSTIME(/SECONDS)-timer' to see how much time has
> elapsed. You should pretty quickly be able to tell if you're actually
> getting a performance hit (which I'd be surprised if you did).
If you have Windows XP and above it will see a hyperthreaded processor as
two processors.
If you have an app that is running a single thread, it's activities will be
roughly divided between the two processors.
If you have an app that has multiple threads that can run concurrently, then
those threads will be divided amongst the processors available.
We run IDL on Windows Server 2003 on a 4-processor system. This shows as 8
processors on the Windows task manager performance tab (the four processors
are hyperthreaded). If we run IDL doing general activity, one processor
generally gets most of the work and the rest do some. If we use IDL doing
things it can multi thread, then all 8 processors bang up to ~90% until it
is done.
Marco
__________________________
Marc Reinig
UCO/Lick
Laboratory for Adaptive Optics
|
|
|
|
| Re: Hyperthreading and IDL? [message #41152 is a reply to message #41142] |
Wed, 29 September 2004 04:23  |
Edd Edmondson
Messages: 50
Registered: January 2003
|
Member |
|
|
Jonathan Greenberg <greenberg@ucdavis.edu> wrote:
> Has anyone done any performance tests looking at IDL 6.0 (or 6.1) under
> hyperthreaded architectures? I see that certain processes max the processor
> out at exactly 50%, which makes me think that I'm getting a 50% performance
> hit just using the hyperthreading enabled. Thoughts? Experience?
It's probably a figment of the way the hyperthreaded CPU appears to the
kernel - as one CPU or as two or whatever. It'd be best to do some sort of
quick benchmark. Use 'timer=SYSTIME(/SECONDS)' before some long routine
and then do a 'print, SYSTIME(/SECONDS)-timer' to see how much time has
elapsed. You should pretty quickly be able to tell if you're actually
getting a performance hit (which I'd be surprised if you did).
--
Edd
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
