XANSYS Message: 20589 [Go back to message list]
[bookmark on del.icio.us]
No rating yet
Rate item:

Subject: Re: win2000 performance
Author: Mark Rodamaker
Date: 2001-01-29 14:01:00

WE have done some testing of NT versus Win2k and have come to the same
conclusions. Assuming you are running in batch, there is a nice summary at
the end of the output file. The 2 OS's seem to take the same amount of CPU
time so the basic system performance is the same. The elapsed or wallclock
times are where the difference occurs with Win2k always slower. Apparently,
Disc I/O is not working as well. This may not be noticeable on some
problems because I/O will be far less if the run allows direct assembly and
the solver is PCG,JCG or ICCG. For nonlinear runs , there will always be a
lot of I/O so the slow I/O will be noticeable.

Mark Rodamaker
MCR Associates
Sunnyvale,California
voice: 408-736-1636

----- Original Message -----
To:
Sent: Monday, January 29, 2001 7:17 AM

> It was probably my email that you are referring to. We tested our new
> Win2000 based software load months ago and compared the performance to
WinNT
> on the same hardware. On some small jobs, no performance difference, but
on
> large jobs, up to 30% more clock time. On the particular test, a
600,000DOF
> model with plasticity and contacts, this performace delta translated to 18
> additional hours of runtime. So I guess my results would seem to match
> yours. I suspect it may be I/O and or driver related and that non-linear
> models might suffer the most if that's true. If it's a driver problem,
etc.
> then there is hope for improvement.

> Kinda irritating to pay more for less, isn't it? ;-)

> Bryan Baskin
> Bell Heligottalovewinderscopter Textron, Inc.
> Texas.

> -----Original Message-----
> From: Forster Daniel [mailto:forsterdaniel@j...]
> Sent: Monday, January 29, 2001 8:55 AM
> To: 'xansys '
> Subject: [xansys] win2000 performance

> I have done benchmarks w/ ansys 5.6.2 on two identical machines, one w/
> win2000 and one w/ winnt and found drastic performance problems on win2000
> (30 to 60% increase in wall time). Some coworkers have not seen this and
in
> fact have used the same benchmark on two separate identical machines and
> found no decrease in performance. I'm surprised that I only found one
email
> in the archive related to performance problems on win2000, maybe something
> is wrong with our win2000 machine. Has anyone else had the opportunity to
> test win2000 on similar hardware that you use for winnt? Our ASD has
> acknowledged the performance problem but didn't give any data yet to
compare
> w/ my numbers.

> On the plus side, we have noticed that win2000 is much better at
> multitasking while ansys is running in batch mode, and may even have
better
> multi processor performance than winnt.

> thanks,

> Daniel Forster
> Product Engineer
> Axle Engineering
> Funk Mfg., Subsidiary of Deere and Company
> Voice (316) 252-3136
> Fax (316) 252-3432

Posts possibly associated with message #20589AuthorDateScore
20572win2000 performanceForster Daniel forsterdaniel@2001/01/29 
20574Re: win2000 performanceBryan Baskin2001/01/29 
20589Re: win2000 performanceMark Rodamaker2001/01/29 
20593Re: win2000 performanceFernand Thomassy2001/01/29 
20601Re: win2000 performanceBryan Baskin2001/01/29 
20611Re: win2000 performanceForster Daniel forsterdaniel@2001/01/30