Author: Christophe Theron
Date: 12:08:11 05/04/00
Go up one level in this thread
On May 03, 2000 at 20:51:04, Robert Hyatt wrote:
>On May 03, 2000 at 19:03:30, KarinsDad wrote:
>
>>On May 03, 2000 at 09:24:34, Robert Hyatt wrote:
>>
>>>On May 03, 2000 at 03:36:45, pavel wrote:
>>>
>>>>Hi,
>>>> there have been a recent debate like "thingie" about crafty "does better
>>>>utilization of computer" then fritz (or most program). In reply to that Bob
>>>>posted that it is not true , however the utilization should be allmost the same.
>>>>And i believed that too. But wasnt sure about it. I downloaded this program
>>>>called "taskinfo", (shareware) and i did a test on crafty and fritz running on
>>>>one cpu under fritz interface. As Bob said the utilization was more or less the
>>>>same, it didnt differ more than 3 to 4% for both programs. It wasnt stable once
>>>>it was 4% better on fritz the other time it was 3% better on crafty. But also
>>>>this differances was also very rare. the most frequent differances was .50%. So
>>>>you can see (Christophe) that crafty_doesnt_utilize_more
>>>>_of_the_computer_resources ;0. To be sure that its the same in all circumstances
>>>>i tried it under winboard (crafty and goliath) and also used other programs
>>>>(exchess,comet,TCB,chessmaster7000). and they are more or less the same.
>>>>just sharing info,
>>>>thanks
>>>>pavel
>>>>
>>>>ps, thus it mean that comp vs comp games are accurate in single cpu. If we
>>>>provide the same hash size, and everything else, wont it be the same for both
>>>>the programs? the only differance would be that in seperate CPU, programs would
>>>>play stronger, but in single cpu they might play comperably weaker.
>>>
>>>
>>>there are hidden problems. IE fritz likes big hash. Using one machine means
>>>it doesn't get it. because in theory, memory should be split 50-50 to be 'fair'.
>>>And this will favor the program that is better at using small hash tables.
>>
>>
>>I don't buy this memory theory.
>>
>>A "fair" test is one of 50% cache for each side (cache memory being more
>>important) and 50% cpu cycles. But other resources should be based on what is
>>required to run the program. For example, if you need 100 MB for one program,
>>well, you'd better have 100 MB + OS requirements + other program requirements.
>>If you need 3GB for egtbs for one program, then you'd better have it.
>>
>>Now, forcing resource allocation is a different issue. And, of course, there
>>will be a resource delay on both sides based on what the other side is doing (if
>>both sides are looking at their egtbs, then there will be a competition for head
>>location on the hard disk). So, you are correct that there will be hidden
>>problems (and a program could be penalized based on how it is coded since it
>>could assume it can grab a large portion of the cache to run in).
>>
>>But I do not believe that spliting main memory 50-50 is required for a fair
>>test.
>>
>>KarinsDad :)
>
>
>(a) why not?
>
>(b) who gets the bigger chunk?
>
>(c) why?
>
>How can it _possibly_ be fair to give one program more hash than another? Or
>give one program more cpu than another?
>
>Cache is immaterial since the thing doesn't context switch that often,
>particularly with ponder=off. So cache simply doesn't factor in, and there is
>no way to control it anyway.
>
>But for the rest???
I totally agree. Equal resources must be affected to each program, as far as
possible.
Cache issues can be tricky, but at least an equal amount of hash tables must be
affected.
If one of the two programs is able to take better advantage of the same amount
of hash tables, it will be rewarded, which is just fair.
This could mean 2 different things:
a) a program should be rewarded if it can avoid a performance drop when only
very little HT are available.
b) a program that can make a good use of very large hash tables should be
rewarded too.
A program able to do both (a) and (b) would of course have a nice advantage.
However I'm not sure that such a program exists.
Christophe
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.