SF 160916 VS Komodo 10.1 long time control

[JJJ]

R. Houdart said H dev is playing with contempt 0 in both stage 3 and rapid!

Such a great score without contempt... I really like the way of how Houdini plays. Frank is gonna love this engines also.

[Lyudmil Tsvetkov]

R. Houdart said H dev is playing with contempt 0 in both stage 3 and rapid!

Such a great score without contempt... I really like the way of how Houdini plays. Frank is gonna love this engines also.

Mon cul, without contempt, -5cps for stage 3, and -7cps in the rapid, if you look at eval at drawn positions.

[Hugo]

to see Stockfish that far in Advance to Komodo 10.1 is what I get too with my matches.
Stockfish made tremendous steps forward.
But this is also a momentary screenshot.
Things can change again
Clemens Keck

You can naturally test however you want, but in science, once you decide on the rules of the tests, you should stick with it until the end. Stopping ealry skews results.

Yes, he should have tested to the end of 100 games. But I have some reserves about this test, draw rate seems too low (openings are not particularly sharp), and +20 -2 is a bit strange. But it is probably an unwarranted skepticism.

[Laskos]

to see Stockfish that far in Advance to Komodo 10.1 is what I get too with my matches.
Stockfish made tremendous steps forward.
But this is also a momentary screenshot.
Things can change again
Clemens Keck

You can naturally test however you want, but in science, once you decide on the rules of the tests, you should stick with it until the end. Stopping ealry skews results.

Yes, he should have tested to the end of 100 games. But I have some reserves about this test, draw rate seems too low (openings are not particularly sharp), and +20 -2 is a bit strange. But it is probably an unwarranted skepticism.

Do you get a draw rate from regular openings on strong hardware and LTC of so low as close to 70%? Is +20 -2 a usual occurrence? It seems to me not the single core performance of Stockfish at bullet time control improved dramatically, but multi-core and a long TC against Komodo. Maybe there was a nasty bug in this respect with later Komodo in these conditions (MP and LTC)? Note that TCEC performance of SF is not that stellar, it is not that far ahead of say Fire.

[mwyoung]

to see Stockfish that far in Advance to Komodo 10.1 is what I get too with my matches.
Stockfish made tremendous steps forward.
But this is also a momentary screenshot.
Things can change again
Clemens Keck

You can naturally test however you want, but in science, once you decide on the rules of the tests, you should stick with it until the end. Stopping ealry skews results.

Yes, he should have tested to the end of 100 games. But I have some reserves about this test, draw rate seems too low (openings are not particularly sharp), and +20 -2 is a bit strange. But it is probably an unwarranted skepticism.

Do you get a draw rate from regular openings on strong hardware and LTC of so low as close to 70%? Is +20 -2 a usual occurrence? It seems to me not the single core performance of Stockfish at bullet time control improved dramatically, but multi-core and a long TC against Komodo. Maybe there was a nasty bug in this respect with later Komodo in these conditions (MP and LTC)? Note that TCEC performance of SF is not that stellar, it is not that far ahead of say Fire.

I am not sure what has happend. But it seems many of us are getting this result at long and fast time controls. Here is a rapid test I am broadcasting and again the same results.

https://www.youtube.com/watch?v=dvEPLU4QwY0

[JJJ]

R. Houdart said H dev is playing with contempt 0 in both stage 3 and rapid!

Such a great score without contempt... I really like the way of how Houdini plays. Frank is gonna love this engines also.

Mon cul, without contempt, -5cps for stage 3, and -7cps in the rapid, if you look at eval at drawn positions.

Nicely seen

[beram]

to see Stockfish that far in Advance to Komodo 10.1 is what I get too with my matches.
Stockfish made tremendous steps forward.
But this is also a momentary screenshot.
Things can change again
Clemens Keck

You can naturally test however you want, but in science, once you decide on the rules of the tests, you should stick with it until the end. Stopping ealry skews results.

Yes, he should have tested to the end of 100 games. But I have some reserves about this test, draw rate seems too low (openings are not particularly sharp), and +20 -2 is a bit strange. But it is probably an unwarranted skepticism.

Do you get a draw rate from regular openings on strong hardware and LTC of so low as close to 70%? Is +20 -2 a usual occurrence? It seems to me not the single core performance of Stockfish at bullet time control improved dramatically, but multi-core and a long TC against Komodo. Maybe there was a nasty bug in this respect with later Komodo in these conditions (MP and LTC)? Note that TCEC performance of SF is not that stellar, it is not that far ahead of say Fire.

I am not sure what has happend. But it seems many of us are getting this result at long and fast time controls. Here is a rapid test I am broadcasting and again the same results.

https://www.youtube.com/watch?v=dvEPLU4QwY0

Or this recent Rapid 15m 3s on 36 cores by Sedat Canbaz
SF 250916 - Komodo 10.1(ct=0) 28 - 12 !
https://sites.google.com/site/computers ... id-36-core

[Dan Cooper]

to see Stockfish that far in Advance to Komodo 10.1 is what I get too with my matches.
Stockfish made tremendous steps forward.
But this is also a momentary screenshot.
Things can change again
Clemens Keck

You can naturally test however you want, but in science, once you decide on the rules of the tests, you should stick with it until the end. Stopping ealry skews results.

Yes, he should have tested to the end of 100 games. But I have some reserves about this test, draw rate seems too low (openings are not particularly sharp), and +20 -2 is a bit strange. But it is probably an unwarranted skepticism.

Do you get a draw rate from regular openings on strong hardware and LTC of so low as close to 70%? Is +20 -2 a usual occurrence? It seems to me not the single core performance of Stockfish at bullet time control improved dramatically, but multi-core and a long TC against Komodo. Maybe there was a nasty bug in this respect with later Komodo in these conditions (MP and LTC)? Note that TCEC performance of SF is not that stellar, it is not that far ahead of say Fire.

100 games between K 10.1 and SF 071616 (K 10.1's "birthdate") I got 71% draws and +23 -6.

300 games between K 10 and SF 052116 I got 80.6% draws (more normal) and +33 -25.

The odd thing is the newer SF doesn't appear any better at beating other engines than K is. It just seems well tuned to specifically beat K.

[Laskos]

to see Stockfish that far in Advance to Komodo 10.1 is what I get too with my matches.
Stockfish made tremendous steps forward.
But this is also a momentary screenshot.
Things can change again
Clemens Keck

You can naturally test however you want, but in science, once you decide on the rules of the tests, you should stick with it until the end. Stopping ealry skews results.

Yes, he should have tested to the end of 100 games. But I have some reserves about this test, draw rate seems too low (openings are not particularly sharp), and +20 -2 is a bit strange. But it is probably an unwarranted skepticism.

Do you get a draw rate from regular openings on strong hardware and LTC of so low as close to 70%? Is +20 -2 a usual occurrence? It seems to me not the single core performance of Stockfish at bullet time control improved dramatically, but multi-core and a long TC against Komodo. Maybe there was a nasty bug in this respect with later Komodo in these conditions (MP and LTC)? Note that TCEC performance of SF is not that stellar, it is not that far ahead of say Fire.

100 games between K 10.1 and SF 071616 (K 10.1's "birthdate") I got 71% draws and +23 -6.

300 games between K 10 and SF 052116 I got 80.6% draws (more normal) and +33 -25.

The odd thing is the newer SF doesn't appear any better at beating other engines than K is. It just seems well tuned to specifically beat K.

A bug in the latest Komodo? SF development is not tuned against Komodo. Were your games on many cores?

[Dan Cooper]

to see Stockfish that far in Advance to Komodo 10.1 is what I get too with my matches.
Stockfish made tremendous steps forward.
But this is also a momentary screenshot.
Things can change again
Clemens Keck

You can naturally test however you want, but in science, once you decide on the rules of the tests, you should stick with it until the end. Stopping ealry skews results.

Yes, he should have tested to the end of 100 games. But I have some reserves about this test, draw rate seems too low (openings are not particularly sharp), and +20 -2 is a bit strange. But it is probably an unwarranted skepticism.

Do you get a draw rate from regular openings on strong hardware and LTC of so low as close to 70%? Is +20 -2 a usual occurrence? It seems to me not the single core performance of Stockfish at bullet time control improved dramatically, but multi-core and a long TC against Komodo. Maybe there was a nasty bug in this respect with later Komodo in these conditions (MP and LTC)? Note that TCEC performance of SF is not that stellar, it is not that far ahead of say Fire.

100 games between K 10.1 and SF 071616 (K 10.1's "birthdate") I got 71% draws and +23 -6.

300 games between K 10 and SF 052116 I got 80.6% draws (more normal) and +33 -25.

The odd thing is the newer SF doesn't appear any better at beating other engines than K is. It just seems well tuned to specifically beat K.

A bug in the latest Komodo? SF development is not tuned against Komodo. Were your games on many cores?

1 core VLTC. If there is a bug it might show itself by running K10.1 v SF 052116. Perhaps I will test that next.