Houdini 5 vs Stockfish 8

[beram]

Have played my first 100 games with Houdini 5
Privat testbook 25 lines, AMD X6 1090T 3200 Mhz
TC 3m+2s, 4CPU, ponder off

Stockfish wins +20 -7 =73 56,5%

Code: Select all

Houdini 5 - SF 8, Blitz 3m+2s                           
1   Stockfish 8 64 POPCNT  +20/=73/-7 56.50%   56.5/100
2   Houdini 5 x64-popc     +7/=73/-20 43.50%   43.5/100

I am underway with 38 games at 5 times longer TC 15m 10s
It seems the fish is doing better with longer time
+7 -1 = 30 with a very high drawpercentage of 78,9% (with same testbook)

My guess for outcome TCEC final 88 draws and +10 - 2 for Stockfish 8

Code: Select all

Houdini 5 - SF 8, Blitz 15m+10s
                           
1   Stockfish 8 64 POPCNT  +7/=30/-1 57.89%   22.0/38
2   Houdini 5 x64-popc     +1/=30/-7 42.11%   16.0/38

[JJJ]

On others test Houdini is doing better against Stockfish. So your test is not enough accurate to me.

[Lyudmil Tsvetkov]

Have played my first 100 games with Houdini 5
Privat testbook 25 lines, AMD X6 1090T 3200 Mhz
TC 3m+2s, 4CPU, ponder off

Stockfish wins +20 -7 =73 56,5%

Code: Select all

Houdini 5 - SF 8, Blitz 3m+2s                           
1   Stockfish 8 64 POPCNT  +20/=73/-7 56.50%   56.5/100
2   Houdini 5 x64-popc     +7/=73/-20 43.50%   43.5/100

I am underway with 38 games at 5 times longer TC 15m 10s
It seems the fish is doing better with longer time
+7 -1 = 30 with a very high drawpercentage of 78,9% (with same testbook)

My guess for outcome TCEC final 88 draws and +10 - 2 for Stockfish 8

Code: Select all

Houdini 5 - SF 8, Blitz 15m+10s
                           
1   Stockfish 8 64 POPCNT  +7/=30/-1 57.89%   22.0/38
2   Houdini 5 x64-popc     +1/=30/-7 42.11%   16.0/38

For some reasons, closely related to SF search and eval specifics, SF excels in the time interval 10-30 min. At blitz, it plays weaker, and also at VLTC.

So, although I also think SF will win the final, it is not claer to me waht will be the score. I guess there will be more than 12 wins, maybe 20.

[Lyudmil Tsvetkov]

Correction, just looked at the first games played at TCEC, and ....
Ooohhh, horror, they are using 12 to 18 moves long book (35 plies), where the game more or less ends.

Whoever made the choice of the book, made a very poor job indeed. The most interesting part of the game, opening and early middlegame, where engines made the most mistakes, is simply left out.

Under those conditions, it is possible that wins will be less, SF will be favoured more, and, absolutely regrettably, we will witness very boring final. (the games have already ended when the engines start thinking, you know)

Also, look where most games end: at move 45-55, far shorter than any menaingful game under those conditions starting at ply 4 or 5. Certainly, a game lasting 100 moves is far more interesting, is not it?

[Guenther]

Have played my first 100 games with Houdini 5
Privat testbook 25 lines, AMD X6 1090T 3200 Mhz
TC 3m+2s, 4CPU, ponder off

Stockfish wins +20 -7 =73 56,5%

Code: Select all

Houdini 5 - SF 8, Blitz 3m+2s                           
1   Stockfish 8 64 POPCNT  +20/=73/-7 56.50%   56.5/100
2   Houdini 5 x64-popc     +7/=73/-20 43.50%   43.5/100

I am underway with 38 games at 5 times longer TC 15m 10s
It seems the fish is doing better with longer time
+7 -1 = 30 with a very high drawpercentage of 78,9% (with same testbook)

My guess for outcome TCEC final 88 draws and +10 - 2 for Stockfish 8

Code: Select all

Houdini 5 - SF 8, Blitz 15m+10s
                           
1   Stockfish 8 64 POPCNT  +7/=30/-1 57.89%   22.0/38
2   Houdini 5 x64-popc     +1/=30/-7 42.11%   16.0/38

For some reasons, closely related to SF search and eval specifics, SF excels in the time interval 10-30 min. At blitz, it plays weaker, and also at VLTC.

So, although I also think SF will win the final, it is not claer to me waht will be the score. I guess there will be more than 12 wins, maybe 20.

20 is mathematically nearly impossible.

Code: Select all

Event        W L  D  N   DRate    Opp1       Opp2         1st Game    Last game  Book moves

TCEC 7 Final 4:7 53  64  82,81%   SF 141214  Komodo 1333  2014.12.16  2014.12.27  8
TCEC 8 Final 9:2 89 100  89,00%   SF 011115  Komodo 9.3x  2015.11.06  2015.11.30  8

Even with a regression to let's say incredible 78% draw rate that would mean 20:2 for SF.

If we assume a more realistic draw rate of may be 85% (if the book was sharpened/ slightly more unbalanced again - ignoring even faster hardware again, which increases draw rates)
that would mean 15 decisive games. From the current test results of perhaps 55% for SF, we could deduct a final result of 12:3 +85 draws, but we don't know exactly what happens at those LTC.

[Guenther]

I did not know that the final already had started and I am also surprised they changed book rules after the last finals, which were fixed to 8 moves length.
IMO it is a pity to use now 11-18! moves long opening lines. All the long time control and hardware which made this interesting now is tied so much too
moves, which were not made by the opponents themselves.
This limits the value of the final a lot.

[JJJ]

I agree, I really don't like the book used here. Way too long. Maybe some opening will be shorter later ?

[Guenther]

I agree, I really don't like the book used here. Way too long. Maybe some opening will be shorter later ?

No idea. I just remember when I did VLTC (4 hours/40 moves) games long ago, I added extremely small opening lines, because I did not want to spoil all those dedicated cpu-time by tiny 'errors' in the book/opening.
That would have been a waste of time... a move which is e.g. analysed for a minute or even a few minutes and looks ok, can still accumulate to a bad move, after several moves with much more thinking time.

[Hai]

Have played my first 100 games with Houdini 5
Privat testbook 25 lines, AMD X6 1090T 3200 Mhz
TC 3m+2s, 4CPU, ponder off

Stockfish wins +20 -7 =73 56,5%

Code: Select all

Houdini 5 - SF 8, Blitz 3m+2s                           
1   Stockfish 8 64 POPCNT  +20/=73/-7 56.50%   56.5/100
2   Houdini 5 x64-popc     +7/=73/-20 43.50%   43.5/100

Last time I saw an AMD X6 1090T 3200 Mhz in the museum!!!

[APassionForCriminalJustic]

Correction, just looked at the first games played at TCEC, and ....
Ooohhh, horror, they are using 12 to 18 moves long book (35 plies), where the game more or less ends.

Whoever made the choice of the book, made a very poor job indeed. The most interesting part of the game, opening and early middlegame, where engines made the most mistakes, is simply left out.

Under those conditions, it is possible that wins will be less, SF will be favoured more, and, absolutely regrettably, we will witness very boring final. (the games have already ended when the engines start thinking, you know)

Also, look where most games end: at move 45-55, far shorter than any menaingful game under those conditions starting at ply 4 or 5. Certainly, a game lasting 100 moves is far more interesting, is not it?

You are way off. The book author is highly credible. And Stockfish is finding no issues with winning, and finding wins. It's a fair Superfinal with a clearly weaker opponent for Stockfish - especially at this time control and huge hardware. The openings also are making a lot of people happy as a lot of the other Superfinals have been rather boring. These openings have led to massively complicated positions, and tactical prowess. I have no idea where you are coming from.