Stockfish 1.7

[Lusakan]

What is credible is that Vas is really down and out cold, you think?

[James Constance]

The windows 32 exe isn't working for me (i'm using windows XP). I've not had problems with any previous version. It installs OK - comes up with the programmer names as per usual - but doesn't start infinite analysis or playing - just dies and freezes the program (shredder classic or fritz).

Is it just me?

I think it may be you, unfortunately.
I've been testing 1.7 against 1.63 (all x86 under XP, / Fritz 11) for ~8 hours now.

Well, I've done a fresh recovery disk install on my netbook, installed all the window updates. Stockfish 1.7 still doesn't work for me. As all previous versions have worked, I guess something critical must have changed. I'll have to try my luck with the next update

[Uri Blass]

Dr Wael,

Did you read this update over at the Rybka Forum?

http://rybkaforum.net/cgi-bin/rybkaforu ... ?tid=16124

Vas is making good progress working on Rybka 4 uci. Online Rybka will come later.

Regards,
Lukas Playchess handles: Rechenschieber, Victor_Kullberg .

If everybody else is making good progress and this knowledge is in the public domain what makes you believe Rybka will remain static? Are'nt you forgetting that Vas was two years ago where these engines are today and he never stood still? Or do you know something we dont?

I believe that Vas made a good progress with rybka but I doubt if he is going to release it.

It is possible that he can make more money by selling it only to one top chess player.

For example if one of topalov and anand pay Vasik 100,000$ per year for not releasing rybka and having a program that is at least 100 elo better than the rest(including the programs that are considered by vasik to be rybka clones) then I see no reason for Vas to release rybka.

It is better to be sure of earning 100,000$ per year and not to release rybka when you are unsure how much money you are going to earn.

Uri

[Milos]

For example if one of topalov and anand pay Vasik 100,000$ per year for not releasing rybka and having a program that is at least 100 elo better than the rest(including the programs that are considered by vasik to be rybka clones) then I see no reason for Vas to release rybka.

It is better to be sure of earning 100,000$ per year and not to release rybka when you are unsure how much money you are going to earn.

Uri, now you're starting to sound like Vincent...

[Eelco de Groot]

The windows 32 exe isn't working for me (i'm using windows XP). I've not had problems with any previous version. It installs OK - comes up with the programmer names as per usual - but doesn't start infinite analysis or playing - just dies and freezes the program (shredder classic or fritz).

Is it just me?

I think it may be you, unfortunately.
I've been testing 1.7 against 1.63 (all x86 under XP, / Fritz 11) for ~8 hours now.

Well, I've done a fresh recovery disk install on my netbook, installed all the window updates. Stockfish 1.7 still doesn't work for me. As all previous versions have worked, I guess something critical must have changed. I'll have to try my luck with the next update

If you have an Atom procesor James, this might be something to do with detection of hyperthreading again? I fear there is no way to turn that off on a netbook? In Shredder Classic maybe you could type "ucidebug" which turns on the debugging function of the GUI and allows you to see which UCI commands are sent to and from the engine before it "dies". Alternatively you might turn on the log function in the UCI options, and maybe the programmers could learn something from a log file. Assuming the engine gets as far as making a logfile...

I'm just making a guess it has something to do with hyperthreading..

Eelco

[Osipov Jury]

I tested Stockfish 1.7 vs Rybka3 (1 CPU, 32 bit) at fixed depth. 412 games from 206 initial positions for every depth. Rybka3 with true depth (+3).

Code: Select all

depth 6: +109 -214  =89 (-91 Elo) time:  292s -  249s
depth 7:  +98 -186 =128 (-75 Elo) time:  549s -  426s
depth 8: +104 -180 =128 (-65 Elo) time:  986s -  769s
depth 9: +108 -157 =147 (-42 Elo) time: 1974s - 1616s

[Gian-Carlo Pascutto]

Tord and his silly understatements. I'm reading +80 ELO over 1.6.

The results seem to be especially good at single core testing, maybe helped by ponder on in Ingo's list.

I'm testing single-core ponder off. But I would suspect the Stockfish team does the same. So I'm puzzled they didn't see big improvement, unless they consider anything under 100 ELO negligible. Harhar!

[Milos]

I tested Stockfish 1.7 vs Rybka3 (1 CPU, 32 bit) at fixed depth. 412 games from 206 initial positions for every depth. Rybka3 with true depth (+3).

Fixed depth testing is just useless. Especially with new more aggressive pruning in SF 1.7.

[M ANSARI]

I am having terrible stability problems when I force 8 threads. Total BSD's ... haven't seen those in a while. I am now testing other engines to rule out some problem that I might be having with my hardware and so far it seems that the problem is only with SF 1.7 when 8 threads are forced. I actually had a problem setting 8 thread and for some reason it went back to 5 CPU's on F12 GUI. But I removed the engine and installed it again and double checked to see if 8 cpu's are on and it seems that this time all CPU's were working ... but again that seems to cause a BSD after only one or two games in the tourney. I have only managed to play 4 games of a 100 game Silversuite_5_1 match against R3 and have given up because of the hard crashes. Hard to say who is stronger on 8 core but the score was +2 wins to R3 and 2 draws. It does seem that search is much more robust but still there is a lot of yo yo in evaluation jumps.

What I noticed most is that the evaluation of SF 1.7 is very close to R3 ... usually within .10 cp. The older SF had a much larger variation. Not sure if it is due to improved search or due to improved evaluation. All in all seems like a strong release and although I have a feeling it is still a bit weaker than R3, it most definitely will catch R3 in the next release if progress continues at the same rate.

[Osipov Jury]

Fixed depth testing is just useless. Especially with new more aggressive pruning in SF 1.7.

Rybka3 has more aggressive pruning than SF 1.7.