Komodo 10.3 released.

[leavenfish]

"Komodo 10.3 on two threads is probably stronger than Komodo 8 on eight threads. In other words, to get the same improvement by upgrading your hardware you might have to spend 30 or 40 times as much money."

For analysis (not game play) would not letting the older engine run longer then be just as good?

I guess however that would depend on how much better K10.3 might be at 'evaluating' random positions for an extended period of time...like say in Chessbase's "Deep Analysis" feature. I've always liked (and I don't have any real data to back this up) Komodo 5.1 (if I remember the version right) for it's 'evaluation' function of random 'non-tactical' positions.

Thanks for the reply!

[mjlef]

"Komodo 10.3 on two threads is probably stronger than Komodo 8 on eight threads. In other words, to get the same improvement by upgrading your hardware you might have to spend 30 or 40 times as much money."

For analysis (not game play) would not letting the older engine run longer then be just as good?

I guess however that would depend on how much better K10.3 might be at 'evaluating' random positions for an extended period of time...like say in Chessbase's "Deep Analysis" feature. I've always liked (and I don't have any real data to back this up) Komodo 5.1 (if I remember the version right) for it's 'evaluation' function of random 'non-tactical' positions.

Thanks for the reply!

Sure, more time can produce a better analysis. Many people run analysis for a day or more on one position. But time is a limited resource. Stronger programs will give a better analysis in the time you have, acting like a faster machine or a machine with more cores. The idea is it is cheaper to buy/use a stronger program than upgrading your hardware.

You can see how elo scales with time here:

http://www.fastgm.de/time-control4.html

This was Komodo 9.3, but the basic curves will be similar (starting at a higher elo with K 10.3, of course). You will notice a drop in elo gain per doubling due to increased draws.

[duncan]

"Komodo 10.3 on two threads is probably stronger than Komodo 8 on eight threads. In other words, to get the same improvement by upgrading your hardware you might have to spend 30 or 40 times as much money."

For analysis (not game play) would not letting the older engine run longer then be just as good?

I guess however that would depend on how much better K10.3 might be at 'evaluating' random positions for an extended period of time...like say in Chessbase's "Deep Analysis" feature. I've always liked (and I don't have any real data to back this up) Komodo 5.1 (if I remember the version right) for it's 'evaluation' function of random 'non-tactical' positions.

Thanks for the reply!

Sure, more time can produce a better analysis. Many people run analysis for a day or more on one position. But time is a limited resource. Stronger programs will give a better analysis in the time you have, acting like a faster machine or a machine with more cores. The idea is it is cheaper to buy/use a stronger program than upgrading your hardware.

You can see how elo scales with time here:

http://www.fastgm.de/time-control4.html

This was Komodo 9.3, but the basic curves will be similar (starting at a higher elo with K 10.3, of course). You will notice a drop in elo gain per doubling due to increased draws.

can you see a time when if a computer is given a choice of black with double time or white it will do better with white?

http://www.open-aurec.com/wbforum/viewt ... f=2&t=2170

[lkaufman]

"Komodo 10.3 on two threads is probably stronger than Komodo 8 on eight threads. In other words, to get the same improvement by upgrading your hardware you might have to spend 30 or 40 times as much money."

For analysis (not game play) would not letting the older engine run longer then be just as good?

I guess however that would depend on how much better K10.3 might be at 'evaluating' random positions for an extended period of time...like say in Chessbase's "Deep Analysis" feature. I've always liked (and I don't have any real data to back this up) Komodo 5.1 (if I remember the version right) for it's 'evaluation' function of random 'non-tactical' positions.

Thanks for the reply!

In standard time control chess, I think that time has already arrived. Having White is worth about forty elo, but the difference between having White and having Black is eighty elo. Top engines at 40/2 hour level gain something like 40 to 50 elo per doubling (depends on number of threads), so White should come out ahead with half the time.

Sure, more time can produce a better analysis. Many people run analysis for a day or more on one position. But time is a limited resource. Stronger programs will give a better analysis in the time you have, acting like a faster machine or a machine with more cores. The idea is it is cheaper to buy/use a stronger program than upgrading your hardware.

You can see how elo scales with time here:

http://www.fastgm.de/time-control4.html

This was Komodo 9.3, but the basic curves will be similar (starting at a higher elo with K 10.3, of course). You will notice a drop in elo gain per doubling due to increased draws.

can you see a time when if a computer is given a choice of black with double time or white it will do better with white?

http://www.open-aurec.com/wbforum/viewt ... f=2&t=2170

[whereagles]

strong correspondence chess players don't usually let engines run overnight.. forcing it to play a couple of moves and evaluating from thereon is usually a better strategy. in this case these speed enhancements will be very welcome

[drj4759]

I think it is inappropriate at this time to announce to the world that "KOMODO RULES". All the rating lists show that it is only number 3 behind Stockfish and Houdini.

It is a shame if you insist to include that Komodo rules in every post you make. You can only do it when you have proof.

[David Xu]

Congrats on the improvement. I'm looking forward to seeing K10.3's performance on the updated rating lists.

I think it is inappropriate at this time to announce to the world that "KOMODO RULES". All the rating lists show that it is only number 3 behind Stockfish and Houdini.

It is a shame if you insist to include that Komodo rules in every post you make. You can only do it when you have proof.

That, my friend, is called a "signature". If you have one, it appears at the end of every post you make. Make sense?

[Adam Hair]

I think it is inappropriate at this time to announce to the world that "KOMODO RULES". All the rating lists show that it is only number 3 behind Stockfish and Houdini.

It is a shame if you insist to include that Komodo rules in every post you make. You can only do it when you have proof.

I see no problem with Larry being enthusiastic about how good Komodo is. After all, he has invested money and a lot of time into its developement. If he made some statement that Komodo was 50 Elo stronger than the latest SF or Houdini without some proof, then that would be shameful. But Larry has never done that.

[drj4759]

What a bad signature it really is. It is out of present reality and should only be used when it is true.

For some like me, such signature is a sore to the eye and an irritant when something else rules.

How would you feel if all engine authors will have their own signature that their respective engine rules? That would be a joke which will cause confusion.

[styx]

Donald you must be fun at parties. There are some 2100 Elo engines, that I think "rule". Take a breath, calm down, think of your weak heart. Gather your energy for important stuff.