Computer Chess Club Archives




Subject: Re: How make Fritz execute brute force search?

Author: Jason Williamson

Date: 18:03:27 07/26/00

Go up one level in this thread

On July 26, 2000 at 19:45:15, leonid wrote:

>On July 26, 2000 at 18:29:36, Torstein Hall wrote:
>>On July 26, 2000 at 18:22:47, leonid wrote:
>>>On July 26, 2000 at 17:26:01, Tom Kerrigan wrote:
>>>>On July 26, 2000 at 16:59:35, leonid wrote:
>>>>>On July 26, 2000 at 13:40:32, Tom Kerrigan wrote:
>>>>>>On July 26, 2000 at 09:18:41, leonid wrote:
>>>>>>>How ask Fritz execute brute force search? I have Fritz 6 but if it is possible
>>>>>>>for some other version (even better DOS version), please say me.
>>>>>>>Recently I went to see Fritz nodes per second performance. Very impressive! Only
>>>>>>>maybe I am missing exact numbers. NPS tend to grow when search is done by brute
>>>>>>>force. This is why I try to find where Fritz numbers stays in real. But Fritz,
>>>>>>>in dispite of its performance, is not exactly open minded piece of software.
>>>>>>>Even its NPS I was able to see only through my Hiarcs 7.32 program.
>>>>>>>Thanks in advance,
>>>>>>My search is selective only because of null-move. I believe this is also the
>>>>>>case with Fritz.
>>>>>>With null move on, my program searches 634k NPS. (BK, short searches)
>>>>>On what hardware do you have 634k? I was very impressed with Fritz numbers only
>>>>>because they were between 220 and 320k on AMD 400Mhz. Your numbers are almost
>>>>>twice as fast.
>>>>Pentium III/800. On a K-6/400 I figure I'd get 350k NPS or so. And that's for BK
>>>>positions; if I did a 5 second run of WAC, I'd get 780k NPS or so.
>>>If I am not missing there something, your numbers are better that the Fritz
>>>have. If you could someday check at what speed Fritz 6 goes on your hardware and
>>>send me the numbers (best with some concret position), it will be very nice. Or
>>>just put them here. If you numbers are so good, like I see them, you should be
>>>really proud to say them in public.
>>What are you talking about? Its more to a chess program than a high NPS! Even
>>when doing a brute force search. Perhaps the Fritz eval is more complex etc.
>>etc. The proof is in the the chess game it plays, not the NPS.
>If you will see all the best programs you will find that biggest part of them
>have surprisingly close NPS, leaving aside few exceptions. When you will write
>your program as amateur (and ever more on C) there are very few chances that
>your NPS will come even close to those numbers. If your NPS is actually not only
>close but even better that the best monsters numbers, you have good chance to
>reach them all later.
>I was amazed by 650K on 800Mhz with a reason. Recently I found that numbers of
>NPS for this computers should be around 2 000 000 NPS for minimax. Only around
>25% of those numbers should be really reachable when all the advanced technics
>of search is used. It give around 400K for 800Mhz Pentium. Raaching 650K is more
>that simply good performance.

On my machine:

Junior 6a roughly 250-300K NPS Rating 2650
Fritz 6a roughly 300-350K nps  Rating 2650
Crafty 17.x 180-200K nps Rating 2550-2600
Hiarcs 7.32 50-75K nps Rating 2600
Little Goliath 2000 300-500k nps Rating 2500-2550
Phalanx roughly 50k nps Rating 2500

and so on.  Arg NPS I see is roughly 100-200k on my machine by the strong
amateurs and around 200K+ with the exception of Hiarcs.  Yet the difference in
speed isn't much between the top amateurs engines and top commercials engines,
so where does the strength lie?  In the evaluation functions.  Don't worry about
nps, it has about as much meaning as BOGOMips do in linux.


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.