Houdini 6 has been released

[BrendanJNorman]

Hi,

I like the way the tactical mode is implemented now ie to use allocated threads.
All or some very nice feature ,I have installed on engine with 50% cores to tactical and 50% to normal run this engine on a position send to IDEA project and use IA in aquarium if any thing looks interesting i then load the normal all cores Houdini see if anything different then if so load a full all cores Tactical to see if it was a tactical solution, then send to IDEA again later run IDEA with Komodo 11.2.2 check any interesting results with SF development and then enter into my own tuned book for Playchess.com it's added a new feature thanks Robert.

Just wondering, did Robert get this idea from Ed's "Speculative Move Principle" in Prodeo 2.2 SMP (which also dedicated one core to the Q3 tactical engine) which essentially does the same thing?

Not saying it's bad, just very curious.

[Leo]

It looks like the learning feature is gone. One of my favorite features.

[Houdini]

Question for Houdart

Found in manual:

The objectively highest level of play is achieved without contempt, i.e. Contempt value 0.

1 - do you urge usage of zero rather than default value when facing SF Dev. Maybe default contempt value is adequate when facing Komodo 11.2.2

2 - Is it wise to allocate one core out of eight for tactical mode during analysis, i.e 7 in normal mode and remaining allocated for tactical mode.

1) Against opponents that are close in strength, contempt zero provides the best result. The difference is small, leaving the default contempt on should cost less than 10 Elo.
2) It will reduce the strength somewhat (also less than 10 Elo), and find slightly faster the tactical solutions. Not a big deal over-all.

[Houdini]

Just wondering, did Robert get this idea from Ed's "Speculative Move Principle" in Prodeo 2.2 SMP (which also dedicated one core to the Q3 tactical engine) which essentially does the same thing?

Not saying it's bad, just very curious.

No, I didn't know about Prodeo 2.2 SMP (apparently from last year).
The Tactical Mode exists since Houdini 3 in 2012, so it's quite an old idea.

[Houdini]

It looks like the learning feature is gone. One of my favorite features.

I've removed the learning feature last year, in Houdini 5.
I wasn't really happy with its implementation and results in Houdini 2/3/4. I may revisit this in the future.

[Nordlandia]

A genuine test position between H6 as white vs K11.2.2 as black

[d]r1b2b1r/ppp2kpp/2n2n2/3pp3/8/8/PPPP1PPP/RNBQ1K1R b - - 0 10

[IWB]

www.inwoba.de is updated now!

Ingo

[Laskos]

www.inwoba.de is updated now!

Ingo

Thanks!

My predictions posted here were off by some 2 points in general offset, but the differences are pretty much there:

The final rating in Ordo Ingo will present tomorrow I get as (error bars are errors of my estimation, not statistical errors of rating):

Code: Select all

Houdini 6:     3351 +/- 2 
Komodo 11.2.2: 3313 +/- 2 
Stockfish 8:   2998 +/- 2

We will see tomorrow if I was right.

I am now tentatively testing the scaling of this Houdini with time control against SF dev, for now it looks promising:

15''+ 0.15''
60''+ 0.60''
Contempt=0

Code: Select all

Games Completed = 400 of 400 (Avg game length = 46.830 sec)
Settings = RR/64MB/15000ms+150ms/M 600cp for 3 moves, D 120 moves/EPD:C:\LittleBlitzer\2moves_v1.epd(32000)
Time = 2709 sec elapsed, 0 sec remaining
 1.  Houdini 6 Pro x64-pext   	191.0/400	90-108-202  	(L: m=1 t=0 i=0 a=107)	(D: r=137 i=25 f=8 s=4 a=28)	(tpm=361.3 d=16.24 nps=1865876)
 2.  Stockfish 170917 64 BMI2 	209.0/400	108-90-202  	(L: m=0 t=0 i=0 a=90)	(D: r=137 i=25 f=8 s=4 a=28)	(tpm=357.2 d=18.05 nps=1547259)


Games Completed = 400 of 400 (Avg game length = 186.268 sec)
Settings = RR/64MB/60000ms+600ms/M 600cp for 3 moves, D 120 moves/EPD:C:\LittleBlitzer\2moves_v1.epd(32000)
Time = 10739 sec elapsed, 0 sec remaining
 1.  Houdini 6 Pro x64-pext   	208.5/400	86-69-245  	(L: m=0 t=0 i=0 a=69)	(D: r=169 i=39 f=8 s=6 a=23)	(tpm=1411.2 d=20.03 nps=1955285)
 2.  Stockfish 170917 64 BMI2 	191.5/400	69-86-245  	(L: m=0 t=0 i=0 a=86)	(D: r=169 i=39 f=8 s=6 a=23)	(tpm=1404.1 d=22.57 nps=1623147)

Maybe will test overnight at 240''+ 2.4''.

[Lyudmil Tsvetkov]

www.inwoba.de is updated now!

Ingo

time for a scientific second run now, including Houdini contempt 0 and latest SF dev.

[Lyudmil Tsvetkov]

www.inwoba.de is updated now!

Ingo

Thanks!

My predictions posted here were off by some 2 points in general offset, but the differences are pretty much there:

The final rating in Ordo Ingo will present tomorrow I get as (error bars are errors of my estimation, not statistical errors of rating):

Code: Select all

Houdini 6:     3351 +/- 2 
Komodo 11.2.2: 3313 +/- 2 
Stockfish 8:   2998 +/- 2

We will see tomorrow if I was right.

I am now tentatively testing the scaling of this Houdini with time control against SF dev, for now it looks promising:

15''+ 0.15''
60''+ 0.60''
Contempt=0

Code: Select all

Games Completed = 400 of 400 (Avg game length = 46.830 sec)
Settings = RR/64MB/15000ms+150ms/M 600cp for 3 moves, D 120 moves/EPD:C:\LittleBlitzer\2moves_v1.epd(32000)
Time = 2709 sec elapsed, 0 sec remaining
 1.  Houdini 6 Pro x64-pext   	191.0/400	90-108-202  	(L: m=1 t=0 i=0 a=107)	(D: r=137 i=25 f=8 s=4 a=28)	(tpm=361.3 d=16.24 nps=1865876)
 2.  Stockfish 170917 64 BMI2 	209.0/400	108-90-202  	(L: m=0 t=0 i=0 a=90)	(D: r=137 i=25 f=8 s=4 a=28)	(tpm=357.2 d=18.05 nps=1547259)


Games Completed = 400 of 400 (Avg game length = 186.268 sec)
Settings = RR/64MB/60000ms+600ms/M 600cp for 3 moves, D 120 moves/EPD:C:\LittleBlitzer\2moves_v1.epd(32000)
Time = 10739 sec elapsed, 0 sec remaining
 1.  Houdini 6 Pro x64-pext   	208.5/400	86-69-245  	(L: m=0 t=0 i=0 a=69)	(D: r=169 i=39 f=8 s=6 a=23)	(tpm=1411.2 d=20.03 nps=1955285)
 2.  Stockfish 170917 64 BMI2 	191.5/400	69-86-245  	(L: m=0 t=0 i=0 a=86)	(D: r=169 i=39 f=8 s=6 a=23)	(tpm=1404.1 d=22.57 nps=1623147)

Maybe will test overnight at 240''+ 2.4''.

so, no clear winner.

Houdini will perform even better at 240 + 2.4, and then, if you decide to increase TC to 1200 + 12, SF will win convincingly again.

so, no clear winner with smaller samples and different TCs.

it would be very interesting to see a very large sample of games at a specific TC.