Author: Christophe Theron
Date: 07:41:57 02/18/02
Go up one level in this thread
On February 18, 2002 at 00:46:41, Uri Blass wrote:
>On February 17, 2002 at 22:40:48, Christophe Theron wrote:
>
>>On February 17, 2002 at 17:58:57, Jonathan Parle wrote:
>>
>>>On February 17, 2002 at 05:31:40, Uri Blass wrote:
>>>
>>>>On February 16, 2002 at 19:52:27, Jonathan Parle wrote:
>>>>
>>>>>On February 15, 2002 at 23:25:23, Lyn Harper wrote:
>>>>>
>>>>>> I've got a Novag Expert, dating back to 1985. A faithful old friend
>>>>>> of mine.
>>>>>> I just now did a little calculation based on the theory about a 70
>>>>>> elo point increase in playing strength for every doubling of clock speed.
>>>>>> If I could get the program out of my Novag Expert and put it on a
>>>>>> floppy disk, it would play at about 2680, right up there with the best
>>>>>> of them. Does this mean there have been no improvements in chess
>>>>>> programming in the lasst 17 years?
>>>>>> I suggest the theory is flawed. The truth is that it works for a
>>>>>> few doublings, then there is a diminishing return.
>>>>>
>>>>>It is a very interesting question, but one that is very hard to answer.
>>>>>Unfortunately Mhz is a horses for courses thing, with there being different
>>>>>processor types. Lyn's Novag Expert, for example, ran on a 6502 processor. Not
>>>>>comparable in clock speed to any PC processor since the 386. And then you have
>>>>>RISC chips, 68000 chips, the 6301Y, Pentiums, Athlons, etc...the list goes on
>>>>>and on. Unfortunately the shear number of hardware combinations and totally
>>>>>different methods of programming make this a question that will always be
>>>>>theorectical. One program might respond "according to theory" by being
>>>>>underclocked and another might totally debunk any theory. Certainly if you could
>>>>>take the program out of a 17 year old dedicated machine and run it somehow on an
>>>>>Athlon 1900XP, you would see an enormous increase in playing strength, but I
>>>>>think it would still be noticeably weaker than other recent programs. Back in
>>>>>the 80's programs were written with one combination of very specific hardware in
>>>>>mind, and consequently they were highly optimised as such. They were actually
>>>>>very efficient, with ELO ratings of over 1900 being achieved on tiny 5Mhz
>>>>>machines with only 32K programs. The PC revolution brought with it significant
>>>>>changes to the way programs could be written. For starters there was much
>>>>>greater processor scaling potential, the ability to incoporate massive opening
>>>>>libraries, large amounts of memory for hash tables (that dedicated machines
>>>>>could only dream about) and the ability to easily and routinely modify reference
>>>>>files used by the program (OK some dedicated machines could do this in a
>>>>>reltively primitive fashion but it was the exception rather than the rule).
>>>>>Today, if you told a programmer they were writing a program for a single, very
>>>>>specific PC, and that PC only, chances are the program would be a little
>>>>>stronger on that PC than a generic program - due to the ability to fine tune the
>>>>>code and the search algorythms. I also believe the reverse also applies. That
>>>>>is, if you could take a first class program of today like Junior or Rebel and
>>>>>somehow port it to run on a 5 Mhz 6502 machine, I believe it would lose a match
>>>>>to a dedicated machine such as Mephisto Polgar.
>>>>
>>>>I do not know about Junior or Rebel but
>>>>I believe that it is not truth for tiger.
>>>>
>>>>palm tiger and chesstiger14.7 are based on the same engine and has the same
>>>>evaluations and search rules if you give it the same hash tables.
>>>>
>>>>I believe that humans learned a lot about chess programs in the last 15 years.
>>>>I believe that if you tell top programmers today to write a program for 5 mhz
>>>>machine they can write a program that is more than 100 elo better than mephisto
>>>>polgar.
>>>>
>>>>Uri
>>>
>>>It would make a very interesting exercise to say the least.
>>
>>
>>Since I have started to adapt the Chess Tiger engine to the Palm (I started in
>>March 2000) I really feel it is an interesting exercise.
>>
>>The Palm is somehow faster than a 5MHz 6502 (it has a 16MHz DragonBall in most
>>models), but it is still much slower than current PCs.
>>
>>On the other hand I can tell you that the Chess Tiger engine has benefited a lot
>>from my efforts to make it run smoothly on the Palms. So the adaptation to the
>>Palm has helped a lot the PC version.
>>
>>
>>
>> Christophe
>
>I do not see why do you need the palm for it.
>
>I think that it is possible to tune the engine for very fast time control on a
>pc.
>When you know the number of nodes per seconds you can use number of nodes as
>your clock because the real clock is not sensitive to times like 1/10000
>seconds.
>
>I think that tuning the program for the palm is more work than it (for example I
>guess that you need to use 16 bit numbers instead of 32 bit numbers)
>
>Uri
Using 16 bits numbers instead of 32 has been no problem for me. The Tiger engine
needs very few 32 bits integers.
The problem was not here actually.
What I have understood by porting Chess Tiger to the Palm is that when the
computing resources are scarce the evaluation function plays a bigger part.
When you have a lot of computing power, evaluation inaccuracies are most of the
time corrected by a deep search.
When you do not, evaluation mistakes just kill you.
So I had to improve Tiger's evaluation.
The impact has been significant on slow computers. On fast computers it also
improved the playing strength, but less significantly.
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.