Komodo 9.2

[lkaufman]

15 elo at fast time control.
Is this more at mid time or long time control ?

We don't have data on that yet. Normally elo gains decline as the time control becomes longer, but there are exceptions.

[lkaufman]

Larry and Mark, are you aware that another Komodo, Komodo Activestate, has been updated to 9.2 too? Are you in touch with them for some coordination? LOL

Such a funny coincidence!

See http://komodoide.com/download/

I never heard of it, so yes, it is a funny coincidence.

[lkaufman]

Against a close rival (of which there is currently only one other than Komodo 9.1) best results will be obtained with a smaller contempt value, perhaps 5.

Hmm....there is a small, but perhaps significant difference between the figure stated here and that in the Official Readme of Komodo 9.2.
Which is more accurate ?
Kindly clarify.

It depends on the meaning of "close". Against another copy of itself, zero is best. Against Komodo 9 or Stockfish 6, I would use 5 for best results. Against latest Stockfish dev, I would use zero. My prediction is that the rating groups which test in default mode will not see much improvement against Stockfish 6 but will see more than 15 elo against the other engines, so the rating gain should be 15 or more with enough games.

[TShackel]

It depends on the meaning of "close". Against another copy of itself, zero is best. Against Komodo 9 or Stockfish 6, I would use 5 for best results. Against latest Stockfish dev, I would use zero. My prediction is that the rating groups which test in default mode will not see much improvement against Stockfish 6 but will see more than 15 elo against the other engines, so the rating gain should be 15 or more with enough games.

Hi Larry,

If you were playing a tournament that included both Stockfish development version and several other engines that are weaker by 100+ points, would you use 0 in anticipation of stockfish or would you use a full 15 (the default) due to the games with the weaker engines knowing that you may lose something against stockfish development version?

Thanks.

Sincerely,

Tim.

[Werewolf]

Hi Larry,

Could you estimate the gain from 9.1 to 9.2 without the contempt? I'm interested in what the gain will be purely for analysis at 10 min per move on a single core. (and also on 24 cores SMP at 30 min / move).

Sorry, I know that's a big ask!
Carl

[lkaufman]

It depends on the meaning of "close". Against another copy of itself, zero is best. Against Komodo 9 or Stockfish 6, I would use 5 for best results. Against latest Stockfish dev, I would use zero. My prediction is that the rating groups which test in default mode will not see much improvement against Stockfish 6 but will see more than 15 elo against the other engines, so the rating gain should be 15 or more with enough games.

Hi Larry,

If you were playing a tournament that included both Stockfish development version and several other engines that are weaker by 100+ points, would you use 0 in anticipation of stockfish or would you use a full 15 (the default) due to the games with the weaker engines knowing that you may lose something against stockfish development version?

Thanks.

Sincerely,

Tim.

The default (15) was chosen precisely for the situation you describe, so I would use that.

[lkaufman]

Hi Larry,

Could you estimate the gain from 9.1 to 9.2 without the contempt? I'm interested in what the gain will be purely for analysis at 10 min per move on a single core. (and also on 24 cores SMP at 30 min / move).

Sorry, I know that's a big ask!
Carl

I think that for blitz chess without using contempt (or let's say using a small value like 5) the gain was close to 15 elo; I didn't want to claim any elo for contempt since its value depends on the choice of opposition. Probably 15 elo in blitz becomes 10 or so at 10 min/move/single core and maybe 7 at 30 min/24cores, but we don't have any data on that.

[Lanzo]

Hi Larry,

Could you estimate the gain from 9.1 to 9.2 without the contempt? I'm interested in what the gain will be purely for analysis at 10 min per move on a single core. (and also on 24 cores SMP at 30 min / move).

Sorry, I know that's a big ask!
Carl

I think that for blitz chess without using contempt (or let's say using a small value like 5) the gain was close to 15 elo; I didn't want to claim any elo for contempt since its value depends on the choice of opposition. Probably 15 elo in blitz becomes 10 or so at 10 min/move/single core and maybe 7 at 30 min/24cores, but we don't have any data on that.

Now that you're in a generous mood, can you lend us the secret to Komodo's impeccable SMP algorithm?

[lkaufman]

Hi Larry,

Could you estimate the gain from 9.1 to 9.2 without the contempt? I'm interested in what the gain will be purely for analysis at 10 min per move on a single core. (and also on 24 cores SMP at 30 min / move).

Sorry, I know that's a big ask!
Carl

I think that for blitz chess without using contempt (or let's say using a small value like 5) the gain was close to 15 elo; I didn't want to claim any elo for contempt since its value depends on the choice of opposition. Probably 15 elo in blitz becomes 10 or so at 10 min/move/single core and maybe 7 at 30 min/24cores, but we don't have any data on that.

Now that you're in a generous mood, can you lend us the secret to Komodo's impeccable SMP algorithm?

Well, I'll just say that it doesn't have much in common with Stockfish or Ippolit or other well-known open-source engines, and that we seem to have improved it slightly today.

[FWCC]

Hello Larry,it seems that Komodo 9.2 picks up the endgame tablebases at a faster rate and with more pieces on the board.In other words I get TB hits much sooner with the board almost "full of pieces" and the hits go deeper than most engines.Can you explain?




FWCC