| A few IDL benchmarck results [message #27340] |
Thu, 18 October 2001 10:53  |
K. Bowman
Messages: 330
Registered: May 2000
|
Senior Member |
|
|
Here are a few results from an IDL code I use for benchmarks. It is a
medium-sized code (~2000 lines altogether). It does no graphics. It
does a fair amount of I/O (netCDF), which uses ~10% of the cpu time.
It typically uses ~100 MB of memory, so it is well outside of cache
sizes. The bulk of the computational time is spent doing interpolation
(indirect indexing of arrays and vectorized multiplication), so it does
not make very efficient use of caches. There are no FOR loops in the
computational part of the code. It is entirely single-precision.
System Clock CPU OS version Time (s)
----------- ------- ------ ---------------------------- --------
Alpha DS20 667 MHz 21264 Tru64 UNIX V5.0A (Rev. 1094) 37
Powermac G4 867 Mhz PPC G4 Mac OS 9.2 under Mac OS X 58
Powermac G4 500 MHz PPC G4 Mac OS 9.1 77
Powerbook G4 400 Mhz PPC G4 Mac OS 9.1 90
Alpha 500au 21164? Digital Unix V4.0 878 108
SGI O2000 180 MHz? IRIX 6.5 IP27 136
SGI O200 180 MHz? IRIX 6.5 IP27 150
SGI O2 IRIX 6.3 IP32 328
Sorry that I don't have all the specs. Some of these machines are so
old I don't remember.
All the calculations are single-threaded. I'm hopoing to re-run them
soon with IDL 5.5 and multi-threading turned on.
We have some 1.7 GHz DP Pentium boxes running Linux. I'll add some
numbers when our sysadmin gets IDL installed.
The PowerMac looks very good in comparison to the much more expensive
Alpha DS20. I was looking forward to a dual processor PowerMac G4 for
about 20% of the cost of the DS20. (NB: due to single-precision
Altivec unit, double-precision codes would not do nearly as well on the
Mac.)
Ken
|
|
|
|
| Re: A few IDL benchmarck results - question [message #27407 is a reply to message #27340] |
Sun, 21 October 2001 11:20  |
Robert Stockwell
Messages: 74
Registered: October 2001
|
Member |
|
|
Interesting.
I can also post some numbers from the different computers I use.
But I have a question.
I have a laptop with 1.2 Ghz Pentium, 512M ram running win2000.
I compared it to an Athlon 1.4 ghz , 512M ram desktop running Redhat 7.0.
Both run IDL 5.4, and execute identical code. No graphics, other than
printing a line about every 5 seconds. It is all floating point
calculations.
Disk access neglibable, and entire process takes about 80 M of ram.
The weird thing is that my 1.2G laptop is a little quicker than the 1.4G
desktop,
in spite of the fact that my sys guy guaranteed that the athlon is screaming
fast
on FP operations (faster than a 1.8Ghz pentium).
An iteration on the 1.2G takes 24 minutes, on the 1.4G it takes 27 minutes.
I was expecting the 1.4G to take about 14 minutes (having been told
how fast the athlon is as compared to a pentium).
Also, time_test2 takes 1.6 seconds on the 1.2Ghz laptop,
and a little over two full seconds on the 1.4Ghz desktop.
So here's the question, is IDL and/or win2000 optimized for a pentium chip?
Why is it so fast?
I do know that the IDLDE is way way better on win2000 than on linux, so are
there compiler optimizations in IDL that only exist on win2000?
Cheers,
bob
PS I don't know if this is relevant, but on the 1.4Ghz linuz box, I pull IDL
over
from another computer to run it on mine. I can't see how that makes a
difference
though since it should be sitting in ram.
K. Bowman <k-bowman@null.tamu.edu> wrote in message
news:181020011253237910%k-bowman@null.tamu.edu...
> Here are a few results from an IDL code I use for benchmarks. It is a
....
> System Clock CPU OS version Time (s)
> ----------- ------- ------ ---------------------------- --------
> Alpha DS20 667 MHz 21264 Tru64 UNIX V5.0A (Rev. 1094) 37
> Powermac G4 867 Mhz PPC G4 Mac OS 9.2 under Mac OS X 58
> Powermac G4 500 MHz PPC G4 Mac OS 9.1 77
> Powerbook G4 400 Mhz PPC G4 Mac OS 9.1 90
> Alpha 500au 21164? Digital Unix V4.0 878 108
> SGI O2000 180 MHz? IRIX 6.5 IP27 136
> SGI O200 180 MHz? IRIX 6.5 IP27 150
> SGI O2 IRIX 6.3 IP32 328
...
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
