komodo wins stage 3

[Milos]

Interesting numbers. - Nevertheless, I do not doubt that Houdini will win the superfinal, as ususal. But the path to it will be exiting!

I the last nTCEC Houdini won against Stockfish 25-23 in the Superfinal. Now there is a 70+ points stronger Stockfish, and a multicore Komodo. Houdini dev. is improved compared to Houdini 3, but let's see how it fares against other two much improved engines.

The latest SF 021113 is on par with H3 with contempt 0 on LTC (I still don't have enough games though to be conclusive, but it seams like that).
So talking about 70+ stronger SF in real world is just dreaming.

70+ points Stronger SF than the one which participated in season 1 and lost 23 to 25 to Houdini 3 in the Superfinal. Well, maybe in TCEC conditions 50 points is more reasonable. When we talk about ELO gains, we must always specify the time control and hardware.

Again you are posting numbers that you just imagined. Please refer to Adam for real numbers.

[Don]

Interesting numbers. - Nevertheless, I do not doubt that Houdini will win the superfinal, as ususal. But the path to it will be exiting!

I the last nTCEC Houdini won against Stockfish 25-23 in the Superfinal. Now there is a 70+ points stronger Stockfish, and a multicore Komodo. Houdini dev. is improved compared to Houdini 3, but let's see how it fares against other two much improved engines.

The latest SF 021113 is on par with H3 with contempt 0 on LTC (I still don't have enough games though to be conclusive, but it seams like that).
So talking about 70+ stronger SF in real world is just dreaming.

70+ points Stronger SF than the one which participated in season 1 and lost 23 to 25 to Houdini 3 in the Superfinal. Well, maybe in TCEC conditions 50 points is more reasonable. When we talk about ELO gains, we must always specify the time control and hardware.

I don't know if this is a factor, but the old Stockfish was quite weak at slow time controls - in fact it sucked. So you have to talk about ELO gain in terms of relevant time controls. Even SF 3 could give Houdini 3 a good game at very long time controls although Houdini was still superior. But it could not touch Houdini 3 at anything near blitz time controls.

It's my feeling that SF has made HUGE progress, but a lot of the tuning has improved it's performance at bullet and blitz time controls much more than at long time controls. SF is starting to act like one of those fast starters now, a new program that excels more at bullet than long TC.

However, just so as not to offend the SF crazies, SF has improved substantially at ALL time controls.

[Laskos]

Interesting numbers. - Nevertheless, I do not doubt that Houdini will win the superfinal, as ususal. But the path to it will be exiting!

I the last nTCEC Houdini won against Stockfish 25-23 in the Superfinal. Now there is a 70+ points stronger Stockfish, and a multicore Komodo. Houdini dev. is improved compared to Houdini 3, but let's see how it fares against other two much improved engines.

The latest SF 021113 is on par with H3 with contempt 0 on LTC (I still don't have enough games though to be conclusive, but it seams like that).
So talking about 70+ stronger SF in real world is just dreaming.

70+ points Stronger SF than the one which participated in season 1 and lost 23 to 25 to Houdini 3 in the Superfinal. Well, maybe in TCEC conditions 50 points is more reasonable. When we talk about ELO gains, we must always specify the time control and hardware.

Again you are posting numbers that you just imagined. Please refer to Adam for real numbers.

What you are talking about? SF regressions showed +50-60 for SF 4 against SF 3, and now +60 over SF 4. If anything 70+ at bullet TC is an understatement.

[OneTrickPony]

By one point.

I see one and a half
I am impressed by Komodo play I have to say. Two stages in a row by big margin and many beautiful games along the way.

I heard that Robert Houdart has added over 50 ELO which means it's almost a sure thing.

The version playing in stage 3 was supposedly already "40-50" ELO stronger. I didn't see him claim he added another 50 points and it seems kinda unlikely.

[kgburcham]

Congratulations and best wishes for the next stage.

Yeah, Houdini will come into its own in the Final stage...then Komodo will certainly need all your best wishes

Yes, I am expecting Houdini to come on strong for stage 4. I heard that Robert Houdart has added over 50 ELO which means it's almost a sure thing.

another upgrade?
So is that two upgrades for this TCEC tournament?

[Don]

By one point.

I see one and a half
I am impressed by Komodo play I have to say. Two stages in a row by big margin and many beautiful games along the way.

I heard that Robert Houdart has added over 50 ELO which means it's almost a sure thing.

The version playing in stage 3 was supposedly already "40-50" ELO stronger. I didn't see him claim he added another 50 points and it seems kinda unlikely.

I heard that he added over 50 ELO to Houdini 3, not an additional 50 ELO to some dev version that is already stronger by 50 ELO.

But even if he has gained 50 ELO over Houidni 3 it would make Houdini 3 a clear favorite. There is a pretty big margin of error even in a tournament like this which consists of a lot of games played by the top players - so nothing has happened so far which would refute the 50 ELO claim.

Having said that, I think what is more likely is that he is getting 50 ELO at blitz time controls and the real gain is substantially less but probably non-trivial. It's even possible that the real gain at these time controls is very little, because one achilles heel of the ippo derivatives is scaling problems. It's very difficult to know if Robert has solved that problem.

None of this really matters. We just wait and see what happens and when some program wins the super-final we can make up all sorts of irrational conclusions as we each see fit

[Milos]

Interesting numbers. - Nevertheless, I do not doubt that Houdini will win the superfinal, as ususal. But the path to it will be exiting!

I the last nTCEC Houdini won against Stockfish 25-23 in the Superfinal. Now there is a 70+ points stronger Stockfish, and a multicore Komodo. Houdini dev. is improved compared to Houdini 3, but let's see how it fares against other two much improved engines.

The latest SF 021113 is on par with H3 with contempt 0 on LTC (I still don't have enough games though to be conclusive, but it seams like that).
So talking about 70+ stronger SF in real world is just dreaming.

70+ points Stronger SF than the one which participated in season 1 and lost 23 to 25 to Houdini 3 in the Superfinal. Well, maybe in TCEC conditions 50 points is more reasonable. When we talk about ELO gains, we must always specify the time control and hardware.

Again you are posting numbers that you just imagined. Please refer to Adam for real numbers.

What you are talking about? SF regressions showed +50-60 for SF 4 against SF 3, and now +60 over SF 4. If anything 70+ at bullet TC is an understatement.

We are talking about performance at TCEC and your constant mentioning of improvement at 5+0.15 on single core in context of TCEC is not even straw man argument it's just noise.

[Laskos]

Interesting numbers. - Nevertheless, I do not doubt that Houdini will win the superfinal, as ususal. But the path to it will be exiting!

I the last nTCEC Houdini won against Stockfish 25-23 in the Superfinal. Now there is a 70+ points stronger Stockfish, and a multicore Komodo. Houdini dev. is improved compared to Houdini 3, but let's see how it fares against other two much improved engines.

The latest SF 021113 is on par with H3 with contempt 0 on LTC (I still don't have enough games though to be conclusive, but it seams like that).
So talking about 70+ stronger SF in real world is just dreaming.

70+ points Stronger SF than the one which participated in season 1 and lost 23 to 25 to Houdini 3 in the Superfinal. Well, maybe in TCEC conditions 50 points is more reasonable. When we talk about ELO gains, we must always specify the time control and hardware.

Again you are posting numbers that you just imagined. Please refer to Adam for real numbers.

What you are talking about? SF regressions showed +50-60 for SF 4 against SF 3, and now +60 over SF 4. If anything 70+ at bullet TC is an understatement.

We are talking about performance at TCEC and your constant mentioning of improvement at 5+0.15 on single core in context of TCEC is not even straw man argument it's just noise.

Maybe you were talking about TCEC irrelevant rating in 18 games, I was talking of SF 20,000 games regressions at 60+0.15 bullet TC.

[velmarin]

http://susanpolgar.blogspot.com.es/2013 ... -tcec.html

Code: Select all

Stage 3 of the world’s strongest computer chess championship TCEC has come to an end. Komodo is the winner of the stage with the amazing 11,5/18, a full 1,5 points ahead of competition. Komodo achieved 6 victories, 11 draws, and lost just one game to prove that its win in the previous stage was not a coincidence.

In a very close battle Bouquet, Houdini, and Stockfish shared the second position with 10,0/18. 
The silver medal of the stage goes to Stage 1 winner Bouquet with best tiebreak. 
The bronze is for the top rated engine in the field Houdini, while 4th place is for Stockfish, which is the only engine to achieve double victory in the stage.

Gull, playing with updated version 2.3, is at sole 5th position with 9,5/18.

The drama of Stage 3 was for the 6th position race, as the top six engines qualify to the next Stage 4 for a hexa round robin. Naum 4.6, Hiarcs 14, and Rybka 4.1 achieved 8,5/18. The tension was kept to the very end with the games Rybka – Gull and the 105 mover Houdini – Hiarcs in progress. In the end, Naum’s upgraded version for this stage had better tiebreak, and so it progresses on, while Hiarcs and Rybka are eliminated until the next season of TCEC.

[lkaufman]

By one point.

I see a margin of one and a half points on the final chart. Why are you trying to make Komodo look bad by saying it only won by one point?

Larry