Ply depth and ELO increments with dedicated chess computers

[appleshampogal]

Now that I have two dedicated chess computers: Fidelity Par Excellence and Mephisto Chess Challenger I have become interested in understanding more about the relationship between how many plies have been searched and the increments in strength per ply. For example, from Spacious Mind's website (which is most superb by the way) I learned that Fidelity Par Excellence has an active level of about 1,832 (or 1,835 based on another source) for 30 seconds average response time. Now, I believe this corresponds to (in many cases) the Par Excellence reaching about 5 plies (6 plies in some cases). From what I've been reading, the jump from about 5 to 6 plies is about 100 elo points additionally. Would anyone here say that is about accurate? What about 7, 8, or 9? Are there any decent ways of calculating this to figure it out? I'm interested in factoring this in with my Mephisto Chess Challenger as well. If the Mephisto has a base strength (active level) of about 2,030, then how could I approximate the elo from subsequent ply increments? If anyone knows please comment!

[Sargon1972]

Hello ,

Well the important thing on chess programs is that how they are programmed, u can say for example give it 1000mhz ,but if it is not good programmed it can reach a elo of 1500 even with 1000mhz

For your Par Excel, the longer the time it have for each move the deeper it will go in his search ,and in +- strength.
For example i have a Par Excel on 16mhz that go now over 2100 elo because some programs like Spracklens,Frans Morsch ,Ed Schroder do react very well on increased speed from 80 till 200 elo points
The Novags of David Kittinger with the same gained speed do not reach this elo jump ,because of the way it is programmed
Like for the Chess Challenger from Frans Morsh it can be tuned up to 40mhz!

In both cases of course the search of depth will go deeper and the program will see more ,so in basis it play better moves and stronger ones
So the only way to increase elo is to give the computers u have more time for a move, ore u set it that it plays his move when he have reached a specific depth , maybe 10 ply , then it depends on the position on the board , if many pieces can be taken the calculate will take longer ,to reach that amount of ply

But in basic it is as i wrote in the begin, it depends how the program is programmed , how many knowledge is inside the program , have it hash and so on.
If u give for example a old program like Boris ,10 it,s original speed ,it will play stronger but it ends somewhere because the program play the moves how it is programmed
If i am right u can from Ed Schroder his site download some programs from him like MM IV , and u can put it in Arena with now a days powerful computers with 2ghz on speed, it will play ,and u will see on the search depth it reached his ply,s and it will go no further,also on the end of each max search u will see really "dummy"moves from the program ,back we go here again, how is the program programmed

I hope u understand it what i wrote and what the point is

Good luck
Kr,Hans

[appleshampogal]

In both cases of course the search of depth will go deeper and the program will see more ,so in basis it play better moves and stronger ones

This is what I'm wanting to know. If Par Excellence at 5mhz and reaching a ply depth of 6 plays at 1,835. What will the ELO be if I give it until 7 ply to make a move, 8, 9, or 10. My question comes down to trying to figure that out mathematically. Do you know?



On another note however. How did you tune up the Chess Challenger to 40 mhz? I had the idea of wanting to enhance my Challenger but I was told the entire thing would have to be reverse engineered. Do you know of anyone who'd be willing to do that for me? And would overclocking it harm the machine? If it wouldn't, I'd like them both to get a boost.

[Fernando]

I think the answer to your question is double faced. As another pal here said, with deeper search -because more speed or time- the better the engine plays, BUT it must be examined what "better" means. It is clear that any program, no matter how simple is, will play better with more ply IF with that extra deepness it can see threats that were not seen before, BUT that are inside the capabilities of the program. Say, a mate which comes at 6 moves, not in three, can be seen if you give the extra deep, BUT a positional situation that requires a more refined programming will not be seen not matter how much deep the simple program goes.
Even it can happens that extra deepness can fool a simple program.
Many years ago I realized that my Champion Chess Challenger tended to play better with random mode because that let the engine to escape hos own limitations and that, sometimes, let him to play better moves.
You see, this a complex, ambiguous issue...

fern

[Erik Stillum]

In the old days it was reckoned that a full ply more would give about 200 ELO. Doubling of speed would give 60-70 points. The speed doubling was tested by SSDF, because the manufacturers made computers with equal hardware, but different software (i.e Scisys Turbostar/Superstar and different versions of Conchess among others).

If you play the computer at 30 instead of 60 seconds/move it would not play 60-70 points worse, because the computers permanent brain guesses your move. As far as I recall Larry Kaufman ment the 60-70 points have to be divided by at least 3 og 4.

Regards Erik

[Larry]

Even it can happens that extra deepness can fool a simple program.

+ 1

I have found over the years that weak programs can often use more
time to find a weaker move. For me, when I discover I am beating a
chess computer, the answer is not to put the machine on longer time, but
to put myself on shorter time.
Fernando, your Champion died, but here is a good replacement:
http://www.ebay.com/itm/VINTAGE-chess-c ... 1364938727
It has a different options menu to the Champion and more strength at
30 secs/move. At 5secs/move it is actually weaker than the Champion.
This is because more source code = underclocking.
The seller will ship to Chile.
L

[HPC]

In 1991, a ply was worth about 200 Elo. The general ply to rating correlation was 4 ply = 1400, 5 = 1600, 6 = 1800 ... However, the base value for 4 ply varies from computer to computer based on how much knowledge is in the position evaluator. Some of the old computers lacked pawn structure knowledge thus they were weaker at a given ply than their counterparts that had pawn structure knowledge.

There was a general paper published in Scientific American in 1990 or 1991 on the subject.

[Cyberchess]

Thanks for pointing this out. I was aware that ELO change calculations were obtained by way of CPU clock speed increases, but wasn’t aware of these calculations based on ply depth, referenced in Scientific American.

The necessarily compact programs used in traditional devoted computers will invariably latch onto an eccentric line that leads to material advantage in lieu of lines following the guidelines of sound positional play. This, I feel, results in CPU speed/ply depth increases being much more effective against other machines/programs, since neither are aware of or do not weight properly the positional and strategic objectives that often prove decisive.

Humans, on the other hand, are far less adept at accurately calculating long, forcing lines, though have learned from experience the cost of, for example, being saddled with a bad bishop, being unable to redeploy forces to defend the kingside in time, or winding up with pieces on “bad squares”.

Happy Experimenting Regards,
John

[ricard60]

that is why programs with 4 ply but with more knowledge win programs with 5 ply from time to time.