Who will win the TCEC Season 8 final?

FriedmannC · Post by **FriedmannC** » Tue Nov 03, 2015 10:30 am

Who do you think will win the great final of this amazing tournament?

Laskos · Post by **Laskos** » Tue Nov 03, 2015 2:49 pm

FriedmannC wrote:Who do you think will win the great final of this amazing tournament?

Although until recently I thought Komodo is the favorite, with the latest fast SF compile sent to Martin I get such a decisive result in direct match at 5'+3'', that I am now of the opinion that SF is the favorite. K 9.2 is with Contempt=0.

120 games at 5 min + 3 secs one core ponder off

SF : K

+41 -18 =61
71.5:48.5
67 +/- 44 ELO points

That is even more skewed than what Ingo gets for IPON at 5+3. Ingo also got that SF performs worse against weaker engines, but it's irrelevant for Superfinal. For Komodo to overcome this huge, larger than 2:1 Win:Loss handicap, it would need some of the following:

1/ K of the Superfinal is much stronger than K 9.2
2/ SF scales really poorly compared to Komodo to longer TC
3/ Lazy_SMP on SF performs really bad on 24 threads
4/ SF loses many games on time

All in all, I voted for Stockfish.

stavros · Post by **stavros** » Tue Nov 03, 2015 5:23 pm

Laskos wrote:
FriedmannC wrote:Who do you think will win the great final of this amazing tournament?
Although until recently I thought Komodo is the favorite, with the latest fast SF compile sent to Martin I get such a decisive result in direct match at 5'+3'', that I am now of the opinion that SF is the favorite. K 9.2 is with Contempt=0.

120 games at 5 min + 3 secs one core ponder off

SF : K

+41 -18 =61
71.5:48.5
67 +/- 44 ELO points

That is even more skewed than what Ingo gets for IPON at 5+3. Ingo also got that SF performs worse against weaker engines, but it's irrelevant for Superfinal. For Komodo to overcome this huge, larger than 2:1 Win:Loss handicap, it would need some of the following:

1/ K of the Superfinal is much stronger than K 9.2
2/ SF scales really poorly compared to Komodo to longer TC
3/ Lazy_SMP on SF performs really bad on 24 threads
4/ SF loses many games on time

All in all, I voted for Stockfish.

i dont expect sf to win even if its better than komodo
the reasons ? think smart

my 100% bet to komodo

Laskos · Post by **Laskos** » Tue Nov 03, 2015 6:43 pm

stavros wrote:
Laskos wrote:
FriedmannC wrote:Who do you think will win the great final of this amazing tournament?
Although until recently I thought Komodo is the favorite, with the latest fast SF compile sent to Martin I get such a decisive result in direct match at 5'+3'', that I am now of the opinion that SF is the favorite. K 9.2 is with Contempt=0.

120 games at 5 min + 3 secs one core ponder off

SF : K

+41 -18 =61
71.5:48.5
67 +/- 44 ELO points

That is even more skewed than what Ingo gets for IPON at 5+3. Ingo also got that SF performs worse against weaker engines, but it's irrelevant for Superfinal. For Komodo to overcome this huge, larger than 2:1 Win:Loss handicap, it would need some of the following:

1/ K of the Superfinal is much stronger than K 9.2
2/ SF scales really poorly compared to Komodo to longer TC
3/ Lazy_SMP on SF performs really bad on 24 threads
4/ SF loses many games on time

All in all, I voted for Stockfish.
i dont expect sf to win even if its better than komodo
the reasons ? think smart
my 100% bet to komodo

I tried to be smart and saw immediately that Mark and Larry will bribe Martin.

Henk · Post by **Henk** » Tue Nov 03, 2015 7:00 pm

I only take this serious if a vote means a bet of 100 euro. Of course money will equally be divided between all the winners and bank gets 10%.

Laskos · Post by **Laskos** » Tue Nov 03, 2015 11:26 pm

After 300 games, it seems previously I got a bit extreme result, now it's tamer:

300 games 5 min + 3 secs SF:K one core ponder off
+88 -52 =160
168:132
42 +/- 27 ELO points

It would indicate a close Superfinal, as Komodo probably does scale better to very long time control.

FriedmannC · Post by **FriedmannC** » Wed Nov 04, 2015 11:59 am

With that ,,Hulk'' hardware, I wouldn't be surprised at all to see an equal match after 100 games, with very few decisive games.

Laskos · Post by **Laskos** » Wed Nov 04, 2015 2:49 pm

FriedmannC wrote:With that ,,Hulk'' hardware, I wouldn't be surprised at all to see an equal match after 100 games, with very few decisive games.

Yes, the possibility of equal match is not negligible. For example if the expected score is +9 -7 =84 for one engine, then there is, according to binomial distribution, 20% to get exactly the expected score +9 -7, and 17% to get an equal match +8 -8.

Meanwhile I left overnight SF against K 9.2 for 100 games at 15 mins + 9 secs, one core ponder off, and the result is:

+22 -13 =65 for Stockfish. I expect a very close Superfinal, as Komodo is probably scaling better.

lkaufman · Post by **lkaufman** » Wed Nov 04, 2015 3:41 pm

I ran off 4500 games on 1 thread at 2' + 1" between the two versions submitted for the final. My result is minus 16 elo for Komodo. Now that SF has lazy smp, I don't think we can count on better scaling with more cores, although I hope Komodo still retains better scaling with more time. So I expect a very close match, with the result probably too close to be meaningful.

JJJ · Post by **JJJ** » Wed Nov 04, 2015 4:26 pm

How strong is the Komodo who will play the final Larry plz ?