stockfish 7

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
gerold
Posts: 10121
Joined: Wed Mar 08, 2006 11:57 pm
Location: van buren,missouri

Re: stockfish 7

Post by gerold » Wed Dec 23, 2015 1:50 pm

beram wrote:
lkaufman wrote:
APassionForCriminalJustic wrote:
whereagles wrote:
APassionForCriminalJustic wrote:Oh wow... so clearly with CCRL's hardware and blitz time control Komodo 9.3 is definitely weaker than Stockfish development of November the 25th (Stockfish 251115 64-bit 4CPU).
What K9.3? I see K9.2 there...
Komodo 9.3 is there... but it is slightly lower than K 9.2. Quite a few games have been played also. I strongly doubt that 9.3 matches Stockfish in blitz anyways with only four cores. So it makes sense to me. Stockfish 6 of course is a dead fish now.
On CEGT 40/20 Komodo on 1 thread after 300 games is 66 elo above Stockfish 6, which is more than the likely gain for Stockfish 7 at this level I think. 40/20 is really about 40/10 on modern hardware due to the way they adjust, which is roughly equivalent to four threads in blitz. So we may have a split decision where Komodo tops the CEGT blitz even after SF 7 comes out while SF 7 tops the CCRL blitz list. Of course all of this can change a lot with more games.
Now after some more weeks things have levelled out and the difference between Komodo 9.3 and Stockfish 6 on 1 core is 34 ELo and not anymore the 66 ELO as Larry mentioned above. The difference between them with 4 cores is even smaller 24
Time for 9.4?

http://www.husvankempen.de/nunn/40_40%2 ... liste.html
I have SF 6 plus 20 over Komodo at long TC.

IWB
Posts: 1539
Joined: Thu Mar 09, 2006 1:02 pm

Re: stockfish 7

Post by IWB » Wed Dec 23, 2015 1:57 pm

Modern Times wrote:... perhaps testing every second Komodo version may make more sense.
Sounds good to me :D ;)

Bye
Ingo

User avatar
Leto
Posts: 2028
Joined: Thu May 04, 2006 1:40 am
Location: Dune

Re: stockfish 7

Post by Leto » Wed Dec 23, 2015 3:38 pm

beram wrote:
lkaufman wrote:
APassionForCriminalJustic wrote:
whereagles wrote:
APassionForCriminalJustic wrote:Oh wow... so clearly with CCRL's hardware and blitz time control Komodo 9.3 is definitely weaker than Stockfish development of November the 25th (Stockfish 251115 64-bit 4CPU).
What K9.3? I see K9.2 there...
Komodo 9.3 is there... but it is slightly lower than K 9.2. Quite a few games have been played also. I strongly doubt that 9.3 matches Stockfish in blitz anyways with only four cores. So it makes sense to me. Stockfish 6 of course is a dead fish now.
On CEGT 40/20 Komodo on 1 thread after 300 games is 66 elo above Stockfish 6, which is more than the likely gain for Stockfish 7 at this level I think. 40/20 is really about 40/10 on modern hardware due to the way they adjust, which is roughly equivalent to four threads in blitz. So we may have a split decision where Komodo tops the CEGT blitz even after SF 7 comes out while SF 7 tops the CCRL blitz list. Of course all of this can change a lot with more games.
Now after some more weeks things have levelled out and the difference between Komodo 9.3 and Stockfish 6 on 1 core is 34 ELo and not anymore the 66 ELO as Larry mentioned above. The difference between them with 4 cores is even smaller 24
Time for 9.4?

http://www.husvankempen.de/nunn/40_40%2 ... liste.html
I highly doubt K9.2 is stronger than K9.3, and looking at the results I see that K9.2 played against Rybka 2.3.2a an engine that is about 450 elo weaker, I would definitely remove those 88 games.

Also I see K9.2 got a 3381 performance against H4 and 3393 performance against Ginkgo 1.3 whereas K9.3 got a 3322 performance against H4 and 3327 against Ginkgo 1.3. K9.2 has 200 games each against those two opponents whereas K9.3 only has 100 games each against those two.

I recommend adding 100 games each to K9.3 against those two opponents. I also think adding at least 50 games each to K9.2 against those two opponents might help because I think the performances it got might be too high.

After that K9.2's rating there should be much closer to K9.3, and eventually when more games are added K9.3 will edge out K9.2, perhaps between 5 to 10 elo.

APassionForCriminalJustic
Posts: 415
Joined: Sat May 24, 2014 7:16 am

Re: stockfish 7

Post by APassionForCriminalJustic » Thu Dec 24, 2015 3:26 am

beram wrote:
lkaufman wrote:
APassionForCriminalJustic wrote:
whereagles wrote:
APassionForCriminalJustic wrote:Oh wow... so clearly with CCRL's hardware and blitz time control Komodo 9.3 is definitely weaker than Stockfish development of November the 25th (Stockfish 251115 64-bit 4CPU).
What K9.3? I see K9.2 there...
Komodo 9.3 is there... but it is slightly lower than K 9.2. Quite a few games have been played also. I strongly doubt that 9.3 matches Stockfish in blitz anyways with only four cores. So it makes sense to me. Stockfish 6 of course is a dead fish now.
On CEGT 40/20 Komodo on 1 thread after 300 games is 66 elo above Stockfish 6, which is more than the likely gain for Stockfish 7 at this level I think. 40/20 is really about 40/10 on modern hardware due to the way they adjust, which is roughly equivalent to four threads in blitz. So we may have a split decision where Komodo tops the CEGT blitz even after SF 7 comes out while SF 7 tops the CCRL blitz list. Of course all of this can change a lot with more games.
Now after some more weeks things have levelled out and the difference between Komodo 9.3 and Stockfish 6 on 1 core is 34 ELo and not anymore the 66 ELO as Larry mentioned above. The difference between them with 4 cores is even smaller 24
Time for 9.4?

http://www.husvankempen.de/nunn/40_40%2 ... liste.html
Dude, where the heck are you getting 34 Elo from? If you are going to make statements then make them accurately. Stockfish 6 is old news. Comparing Komodo 9.3 to it is a little ridiculous. Here is a real result from the CEGT website page. The time control is 40\20:

Stockfish 6.0 x64 1CPU - 3182 100 + 29 = 63 - 8 60.5 % 3255

Now yes these are only 100 games total. My rating calculator tells me that we have a 74+ Elo advantage over Stockfish 6. Needless to say, a 60.5% percent score is quite the little spanking.

It should be noted that the four core results are problematic. I will copy and paste here what Larry has written concerning Komodo 9.3:

"The selectivity value was raised because one program change made it possible to be
more selective, and 131 tested as the best value in bullet chess. It is possible
that it is too high for normal levels, but there is no evidence for this.

My view is that Komodo 9.3 is a clear improvement over 9.2 on one thread and on
many threads, but it now seems that the MP changes that helped on many threads were
a small negative on 4 threads, so it's not so clear that 9.3 is stronger than 9.2
on 4 threads. We have already fixed this problem in our latest dev version.

Larry"

Has anyone even bothered to look at the result differences between Komodo 9.3 and 9.2 SINGLE CORE?! I'm seeing +20 in 9.3s favour. Also below is 9.2s score versus SF 6. Not quite as good as 9.3 is it?!

Stockfish 6.0 x64 1CPU - 3182 100 + 26 = 55 - 19 53.5 % 3206

Other than some SMP issues now fixed for lower cores like 4, 9.3 is absolutely stronger. There is far too much hype and paranoia around here.

stavros
Posts: 136
Joined: Tue Dec 02, 2014 12:29 am

Re: stockfish 7

Post by stavros » Sun Dec 27, 2015 8:22 pm

ok stockfish 7 beta 1 is out ,its time for some tests

Post Reply