Leto wrote:How much of a time handicap does Gideon need to play stronger (let's say a score of at least 55%) than Komodo 9.2?
It depends on Gideon's time control. For Gideon on one core at 40/2 minutes the handicap is a factor of 60, at 40/4 hours the handicap factor is close to 1,000. Depends on hardware too, similarly to time control.
So Gideon at 40/2 minutes would be stronger than Komodo 9.2 at 40/1 seconds? Do you have a match result that would support that?
I know on the first post Gideon scored 45% when Komodo 9.2 was at 40/2 seconds, but to get to 55% from 45% would indicate a swing of 72 elo by my calculations which would be a big swing in my opinion.
Leto wrote:How much of a time handicap does Gideon need to play stronger (let's say a score of at least 55%) than Komodo 9.2?
It depends on Gideon's time control. For Gideon on one core at 40/2 minutes the handicap is a factor of 60, at 40/4 hours the handicap factor is close to 1,000. Depends on hardware too, similarly to time control.
So Gideon at 40/2 minutes would be stronger than Komodo 9.2 at 40/1 seconds? Do you have a match result that would support that?
I know on the first post Gideon scored 45% when Komodo 9.2 was at 40/2 seconds, but to get to 55% from 45% would indicate a swing of 72 elo by my calculations which would be a big swing in my opinion.
Yes, the doubling of Komodo from 40/1s to 40/2s is much larger than 70 ELO points. So yes, Gideon 40/2 minutes is stronger than Komodo 40/1 second (if Komodo is testable at this time control).
fern wrote:CM 3000 is a very good piece of programming. Some atributes it to Kittinger, some other to the Spracklen. In any case play a very strong game in current machinery.
I dare to say any one of these old DOS engines circa 1990 are, these days, 2200 or so level players.
Heard that Crassus is going to massacre in Carrhae regards
Fern
I can tell you that the engines Don and I wrote around 1990 or so, running on my then state of the art 25 MHz 486, were definitely of human master (say at least 2300 FIDE or so) strength, and in rapid chess (game 30' or so) scored decently against Grandmasters in the Harvard Cup events.
This list http://www.anacadigital.com/dedicados/listaelo1992.html
is consistently inflated compared to SSDF list by 150 ELO points. http://ssdf.bosjo.net/long.txt
You say that 2300 FIDE is a realistic ELO for a top 1991-2 engine on PC and 68030 (Motorola)? Then, the first list is correct (it has 6 thousands games against humans) and SSDF is deflated in this range by 150 ELO points. Interesting. When comparing to humans, I need faithful anchor ratings.
Several of the machines on the anacadigital list were officially rated by USCF over 48 tournament games against humans with money prizes for success. The four I remember were all higher on the USCF list by 205 to 272 points. However USCF ratings at the time were about a hundred elo above FIDE ratings. My recollection is that most national European ratings, especially German, Dutch, and Swedish, were about a hundred elo BELOW FIDE ratings back then. So that list was probaby reasonably accurate relative to European ratings at the time, which were probably much more common back then for amateur players than FIDE ratings. In interpretating and comparing the computer vs computer lists of today with human rating lists, it is of course necessary to apply the contraction factor, which you estimate at 70% (my own estimate in the past was 75%, pretty close agreement). With all of this information, perhaps you can estimate what the "true" rating of Komodo 9.2 (on say 4 cpus) would be against top humans today. Probably the numbers shown on the CCRL/CEGT lists are too high, even though nuimbers shown for old engines are too low, due to this 70% factor.
Leto wrote:How much of a time handicap does Gideon need to play stronger (let's say a score of at least 55%) than Komodo 9.2?
It depends on Gideon's time control. For Gideon on one core at 40/2 minutes the handicap is a factor of 60, at 40/4 hours the handicap factor is close to 1,000. Depends on hardware too, similarly to time control.
So Gideon at 40/2 minutes would be stronger than Komodo 9.2 at 40/1 seconds? Do you have a match result that would support that?
I know on the first post Gideon scored 45% when Komodo 9.2 was at 40/2 seconds, but to get to 55% from 45% would indicate a swing of 72 elo by my calculations which would be a big swing in my opinion.
Yes, the doubling of Komodo from 40/1s to 40/2s is much larger than 70 ELO points. So yes, Gideon 40/2 minutes is stronger than Komodo 40/1 second (if Komodo is testable at this time control).
Would you be interested in testing that out? I'd be interested in seeing the games from the match if Gideon at 40/2 minutes manages to beat Komodo 9.2 at 40/1 seconds. I have a feeling that Gideon wouldn't win even with that handicap, probably score 48%.
Leto wrote:How much of a time handicap does Gideon need to play stronger (let's say a score of at least 55%) than Komodo 9.2?
It depends on Gideon's time control. For Gideon on one core at 40/2 minutes the handicap is a factor of 60, at 40/4 hours the handicap factor is close to 1,000. Depends on hardware too, similarly to time control.
So Gideon at 40/2 minutes would be stronger than Komodo 9.2 at 40/1 seconds? Do you have a match result that would support that?
I know on the first post Gideon scored 45% when Komodo 9.2 was at 40/2 seconds, but to get to 55% from 45% would indicate a swing of 72 elo by my calculations which would be a big swing in my opinion.
Yes, the doubling of Komodo from 40/1s to 40/2s is much larger than 70 ELO points. So yes, Gideon 40/2 minutes is stronger than Komodo 40/1 second (if Komodo is testable at this time control).
Would you be interested in testing that out? I'd be interested in seeing the games from the match if Gideon at 40/2 minutes manages to beat Komodo 9.2 at 40/1 seconds. I have a feeling that Gideon wouldn't win even with that handicap, probably score 48%.
A test like that is totally meaningless. I suspect that on many GUIs Komodo would simply forfeit most of the games on time at 40 moves in one second. The results would depend almost totally on the efficiency of the GUI and on the setting of "Overhead ms" parameter. Not very interesting. Anything below something like 40/4" or so is just not a test of the engine but of the gui.
Sure it is so.
Don was engaged in gambit Kasparov program I think, you too, right?
Gambit is very entertainning and I wonder how strong it is.
Sure you know...
Fern
fern wrote:Sure it is so.
Don was engaged in gambit Kasparov program I think, you too, right?
Gambit is very entertainning and I wonder how strong it is.
Sure you know...
Fern
Yes, Don and I worked together on the engine for Kasparov's Gambit. I think all of our engines of the early 1990s were close in strength, roughly FM strength on a fast pc from that period. Probably strong GM level on an I7 computer today, if it even runs on it.
It run with D-Fend and run OK, strong and funny with Mr Kasparov coming and saying those alarming sentences like "looks like you are not in the right track now..."
I miss that kind of extra fun in current programs, solely concentrated in engine strength.
