Houdini ?

[JAMerolle]

Does anyone know the projected release date of the next Houdini? Any feedback would be great. I project the next release of this engine will jump over R4.

[gerold]

Does anyone know the projected release date of the next Houdini? Any feedback would be great. I project the next release of this engine will jump over R4.

Some time ago Robert said in Nov.or Dec.
In all tests i have seen in Bliz T.C. Houdini is stronger than R4 now.
I have only seem one rating list that has many games and longer T.C.
It shows the two being the same elo. +- 10 elo.

[JAMerolle]

Houdini was smoked in a head to head match against R4 in a 90 30 match held by Martin T. about 2 month ago. * I think it came out to an 8 point lead by R4.

Therefore I think there is still a considerable distance between the two which I acspect to closed and overtaken by Houdini . Hay but thanks for the feedback.

* http://www.talkchess.com/forum/viewtopi ... =&start=90

[gerold]

Houdini was smoked in a head to head match against R4 in a 90 30 match held by Martin T. about 2 month ago. I think it came out to an 8 point lead by R4.

Therefore I think there is still a considerable distance between the two which I acspect to closed and overtaken by Houdini . Hay but thanks for the feedback.

As you know a few games does not prove one stronger than the other.
Also with just a few set openings things can make a diff.
I have seen many matches of 20 games played between these two
at 6/3 TC where Houndini is 20 elo stronger. This does not mean
Houndini is stronger over all. To few games with to few openings.

I do agree that if Robert can make a few improvements that Houndini
can be stronger than R4 if Vas does not improve R4 any.

[Houdini]

Does anyone know the projected release date of the next Houdini? Any feedback would be great. I project the next release of this engine will jump over R4.

Joseph,

I hope to be able to release Houdini 1.1 around mid-December.
More details about new features will follow later.

Cheers,
Robert

[mwyoung]

I also have Houdini better then Rybka 4 in my rating list.

I do not test with blitz time controls. So Rybka 4 did not have any time trouble issues in my games. I test with X/40moves.

It is a myth that Rybka 4 problems are due to time trouble issues. I have played many games with both Rybka and Houdini. What I see is Rybka 4 & Houdini are about equal in the middle game, but Houdini is better in the endgame. I see many games where Rybka thinks it is much better by its eval, only to draw or lose the endgame to Houdini.



1 Houdini 1.03a w32 4_CPU 2464 917
2 Houdini 1.02 w32 4_CPU 2461 145
3 Stockfish 1.9.1 JA 2450 75
4 Deep Rybka 4 w32 2445 1297
5 Stockfish 1.9 JA 2444 33

[Uri Blass]

I also have Houdini better then Rybka 4 in my rating list.

I do not test with blitz time controls. So Rybka 4 did not have any time trouble issues in my games. I test with X/40moves.

It is a myth that Rybka 4 problems are due to time trouble issues. I have played many games with both Rybka and Houdini. What I see is Rybka 4 & Houdini are about equal in the middle game, but Houdini is better in the endgame. I see many games where Rybka thinks it is much better by its eval, only to draw or lose the endgame to Houdini.



1 Houdini 1.03a w32 4_CPU 2464 917
2 Houdini 1.02 w32 4_CPU 2461 145
3 Stockfish 1.9.1 JA 2450 75
4 Deep Rybka 4 w32 2445 1297
5 Stockfish 1.9 JA 2444 33

I think that 32 bits help more houdini than rybka because Rybka is not optimized for 32 bits(In other words Rybka could be 20% or 30% faster
in 32 bits in case that Vas cared about the right optimizations for 32 bits).

[ernest]

I think that 32 bits help more houdini than rybka

I agree!
Very easy to test, too: take 1-cpu, same hash (for reproducibility) and compare the infinite analysis of the start_position (or other) obtained with the 64-bit and the 32-bit versions.
Houdini x64/w32: x1.28
Rybka x64/w32 : x1.7

[JAMerolle]

Thank you for the feedback I look forward to seeing your net release. Your work has created a stir in the community.

[mwyoung]

Don't know about 64bit, other then the other rating list testing with 64 bit. From what I have seen they have Houdini better also by about the same margin.

Like I said, Houdini is better because of the evaluation function. Rybka 4 misevals endgames very badly. I see Rybka 4 eval endgames at around +2, and Houdini scores it even or better for the other side with Houdini proving its eval is correct by winning or drawing the game. A 20% speed up is not going to correct this problem.

I have both programs, but I use Houdini 95% of the time for analysis because it is just better then Rybka 4.

I am sure by mid December Houdini will make Rybka 4 totally obsolet, if Rybka 4 does not impove.