Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SSDF rating list: two interesting observations

Author: Ernst A. Heinz

Date: 10:46:21 03/05/98

Go up one level in this thread


On March 05, 1998 at 12:29:24, Albert Silver wrote:

>On March 04, 1998 at 08:39:33, Robert Hyatt wrote:
>
>>On March 04, 1998 at 01:15:54, Jouni Uski wrote:
>>
>>>
>>>>This has to be operator error, because there is *nothing* in win95 that
>>>>will cause this behavior, other than setting hash tables a little too
>>>>big
>>>>and causing swapping...
>>>>
>>>
>>>It is not any operator error. Specially in endgames Genius in dos (/X)
>>>mode is often 2X faster, than windows version even with about same
>>>hash sixe.
>>
>>I don't want to get into a heated argument here, but my background is
>>operating system development, and I've been in that environment for
>>almost
>>30 years, and I can't think of a single reason why a program would run
>>2x
>>slower under win95, if you discount the paging problem.  And if you
>>discount
>>the problem of having multiple applications running and sharing cpu
>>resources.
>>
>>Other than those two things, there is *nothing* I can think of that
>>would
>>cost you 2x.  You ought to run the windows resource monitor to see what
>>is
>>going on, because I have never seen this outside of the two exceptions
>>mentioned above...
>
>Well, I can't comment on Genius 5 as I don't have it, but I have noticed
>a similiar (80% difference on the average) performance hit when running
>Mchess 7.01 in DOS as opposed to Win95. I can assure that it is not due
>to any TSRs or other software running in the background as if I call up
>the special Rebel9 windows shortcut it runs almost identically in speed
>as when run in DOS. I have no explanation for it but I can give you test
>suite results with the resource monitor's results as well for your
>examination.
>
>                               Albert

A plausible explanation for this strange behavior might be that task
switches under Windows NT/95 badly thrash some CPU internals (caches,
pipelines, branch prediction units etc.) -- I remember having read
some tests which suggest exactly this. Some integer-bound applications
therefore seem to run much slower under Windows NT/95 than under DOS.

=Ernst=



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.