If you do a full install of the developer tools, you get a hardware
control panel
in your system preferences. You can use this to turn off a second CPU
and thus
get reliable 1 CPU v.s 2 CPU benchmarks.
-Nathan
On Jan 23, 2004, at 12:51 AM, Peter Klaver wrote:
Hi everyone,
I didn't do any profiling. This only came up as I was putting
together an XServe G5. I looked into Shark but it wasn't obvious [to
me] how to use it without a large time investment.
The code is a Fortran 77/90 code compiled via xlf_r -O5 -qstrict
code.f
Are you not using more than 1 cpu on average when you do the single
job? I've noticed that when you look at top while running xlf
executables, they can take more than 100% occasionally. And I don't
mean just a short blib at 103% but something like 140% for dozens of
seconds. This can also happen when you have not given any compiler
options to auto-detect parallel threads or anything like that. If that
top output is really correct, you may use something like 1.2 cpus on
one job on average. So naturaaly you'll get less gain when going to
two jobs.
To test this, just run your code and stare at 'top -u' in another
window for a while.
If this is the case it would explain your results very well. I don't
think the G5 is twice as fast as the Opteron for most codes. Your
single-job result may just have been over-optimistic.
kind greets,
Peter
--
+++++++++++++++++++++++++++++++++++++++
The requirements said: Windows 2000 or better.
So I got a Macintosh.
Peter Klaver
private:
Noordeinde 28
2311 CE Leiden
071-5137357
at work:
Delft University of Technology
e-mail email@hidden
room 222, 015-2789518
Rotterdamse Weg 137
2628 AL Delft
the Netherlands
_______________________________________________
scitech mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/scitech
Do not post admin requests to the list. They will be ignored.
(The opinions expressed in this email are the opinions
of Nathan Strange alone and do not neccesarily represent
the official position of any orginization including JPL, NASA,
and CalTech.)
_______________________________________________
scitech mailing list | email@hidden
Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/scitech
Do not post admin requests to the list. They will be ignored.
References:
>G5 bandwidth (From: "Sean C. Garrick" <email@hidden>)
