Official Release of Ethereal 11.75, supporting MultiPV

[Guenther]

It's probably better if you could set the perfect setup for Ethereal 11.75 under Linux/cutechess and then if it's running perfect, my "obviously bad setup" is causing the problem.

Why should I? I'm not interested in testing Ethereal like you seem not interested in understanding what processor cores and concurrency mean.

Bye.

I realize now that we already went through this, more than two years ago!
viewtopic.php?f=6&t=65310

Nothing seems to help here. No wonder, 'chessowl' is just infamous for spreading all kind of SF clones.
Edit: Note that the so called 'Chess Engines Diary' is even worse...

And still he could not resist now to post his latests 'result' there:
http://chessowl.blogspot.com/2019/11/et ... gines.html

Code: Select all

 Ethereal 11.75 vs. Top Chess Engines

 Rank  	Chess Engine 	 	 ELO   	 Games  Score% 	Draw% 	Author
 1    Komodo 11.2.2 x64 	 4084  	 100   	99.00 	0.00 	Dailey, Kaufman, Lefler
 2    Fire 7.1 x64 	 	 4084  	 100   	99.00 	0.00 	Norman Schmidt
 3    Shredder 13 x64 	 	 4012  	 100   	98.50 	1.00 	Stefan Meyer-Kahlen
 4    Houdini 6.02 x64 		 4012  	 100   	98.50 	3.00 	Robert Houdart
 5    Xiphos 0.6 x64 	 	 3920  	 100   	97.50 	3.00 	Milos Tatarevic
 6    Rofchade 2.2 x64 		 3920  	 100   	97.50 	1.00 	Ronald Friedrich
 7    Equinox 3.30 x64 	 	 3888  	 100   	97.00 	0.00 	Colli, Rocchi
 8    Schooner 2.1 x64 		 3835  	 100   	96.00 	0.00 	Dennis Sceviour
 9    Laser 1.7 x64 	 	 3814  	 100   	95.50 	1.00 	Jeffrey An, Michael An
 10   Defen 2.2 x64 	  	 3794  	 100   	95.00 	0.00 	Can Catin, Dogac Eidenk
 11   Nemorino 5.00 x64 	 3794  	 100   	95.00 	0.00 	Christian Gunther
 12   Andscacs 0.95 x64 	 3777  	 100   	94.50 	1.00 	Daniel Jose Queralto
 13   Rubi 1.6 x64 	 	 3732  	 100   	93.00 	0.00 	Andreas Matthias
 14   Demolito 2019-11-03 x64 	 3732  	 100   	93.00 	0.00 	Lucas Braesch
 15   Hannibal 1.7 x64 	 	 3732  	 100   	93.00 	0.00 	Sam Hamilton, Edsel Apostol
 16   Nirvana 2.4 x64 	 	 3684  	 100   	91.00 	0.00 	Thomas Kolarik
 17   Senpai 2.0 x64 	 	 3663  	 100   	90.00 	0.00 	Fabien Letouzey
 18   Texel 1.07 x64 	 	 3627  	 100   	88.00 	0.00 	Peter Osterlund
 19   ICE 4.0.853 x64 	 	 3604  	 100   	86.50 	1.00 	Thomas Petzke
 20   Arasan 21.4 x64 		 3516  	 100   	79.50 	1.00 	Jon Dart
 21   Vajolet 2.2.8.0 x64 	 3500  	 100   	78.00 	0.00 	Marco Belli
 22   Ethereal 11.75 x64 	 3278  	 2200    6.59 	0.60 	Andrew Grant

[RubiChess]

I realize now that we already went through this, more than two years ago!
viewtopic.php?f=6&t=65310

Nothing seems to help here. No wonder, 'chessowl' is just infamous for spreading all kind of SF clones.

Indeed. I wonder what goes wrong with guys like him.

[dannyb]

thanks for this new release Andrew and I wish you the best of health.

[AndrewGrant]

A year ago, I was using Arena 3.5 as my primary chess engines tournament program using a maximum of 7 instances to run a tournament. I got the inspiration to use 32 concurrencies with cutechess when Ethereal became very strong and Andrew revealed that he was using 32 bit concurrencies in cutechess to run million games to test Ethereal. I was intrigued, tested it myself and was blown that it could even run when I increased it to 64 concurrencies under Linux. This was a huge resource saver. Instead of using five 5 core AMD computers to run tournaments for rating list publication, I could just use one 4 core computer which is what I am having now.

I am also surprised that Andrew is not fully aware that Ethereal with Windows binary could run on Linux where WINE is installed. Perhaps, I am the only one running chess engine tournaments with all Windows binary executables under Linux. The best way to verify this is for the unbelievers to test it for themselves and then share their experiences.

The proof that Ethereal is working under cutechess and Linux is to visit my website chessowl.blogspot.com and download all the gauntlet pgn's then look the contents. All other primary engines running under the gauntlet also run with cutechess and Linux including all the Stockfish and Clones. The pgn's are unaltered and you could see what cutechess is commenting for each move. All of these engines are Windows binaries with .exe extension which you can also download in the tournament score board links.

I play with 32 concurrency because I have 32 cores in a single machine. Really I should play 30 concurrency, as one thread/core should be reserved for The OS, Cutechess, and my Python application that runs everything.

You are not saving resources. You are reducing the effective TC (at best, at worst you butcher the results). You can 8x as many processes as you should, and as a result each engine gets (at best, at worst some get none) 1/8th the time you actually intend to give. Again, you are not saving resources.

I'm still not confident that Ethereal works under WINE ever, I imagine your 11.73 games were either with Linux, or had connection stalls that you did not notice, because there is nothing between 11.73 and 11.75 that would change how WINE works. Nothing! I added a UCI option, I added some normal Ethereal code, and I deleted a mutex. Those things would not impact WINE. Adding windows specific things would impact WINE. Adding new libs would impact WINE. Working with the existing code base should have no impact.

Further more, I don't understand why you won't just run a Linux version. Linux works perfectly fine, and I've got ~76 million games that show that. You've got a few hundred that show problems with WINE. I don't support WINE. I provide Windows and ARM binaries, and I provide source for those who make use of Linux and at-home compilation.

I'm not going to attack your testing process with the same vigor as others; but your process is bad, and generates tainted results. I would be much more interested in your rating list if I knew the process was solid at a fundamental level. But at this point, between WINE and the poor testing setup, I don't know whether to invest the time to setup WINE myself and work on Ethereal, or just hand-wave this thread as a hardware problem.

[Raphexon]

32 concurrency on 4 cores is next level.

Even defending it. I'll be nice and call you an idiot.

[lucasart]

It's probably better if you could set the perfect setup for Ethereal 11.75 under Linux/cutechess and then if it's running perfect, my "obviously bad setup" is causing the problem.

Why should I? I'm not interested in testing Ethereal like you seem not interested in understanding what processor cores and concurrency mean.

Bye.

I realize now that we already went through this, more than two years ago!
viewtopic.php?f=6&t=65310

Nothing seems to help here. No wonder, 'chessowl' is just infamous for spreading all kind of SF clones.
Edit: Note that the so called 'Chess Engines Diary' is even worse...

And still he could not resist now to post his latests 'result' there:
http://chessowl.blogspot.com/2019/11/et ... gines.html

Code: Select all

 Ethereal 11.75 vs. Top Chess Engines

 Rank  	Chess Engine 	 	 ELO   	 Games  Score% 	Draw% 	Author
 1    Komodo 11.2.2 x64 	 4084  	 100   	99.00 	0.00 	Dailey, Kaufman, Lefler
 2    Fire 7.1 x64 	 	 4084  	 100   	99.00 	0.00 	Norman Schmidt
 3    Shredder 13 x64 	 	 4012  	 100   	98.50 	1.00 	Stefan Meyer-Kahlen
 4    Houdini 6.02 x64 		 4012  	 100   	98.50 	3.00 	Robert Houdart
 5    Xiphos 0.6 x64 	 	 3920  	 100   	97.50 	3.00 	Milos Tatarevic
 6    Rofchade 2.2 x64 		 3920  	 100   	97.50 	1.00 	Ronald Friedrich
 7    Equinox 3.30 x64 	 	 3888  	 100   	97.00 	0.00 	Colli, Rocchi
 8    Schooner 2.1 x64 		 3835  	 100   	96.00 	0.00 	Dennis Sceviour
 9    Laser 1.7 x64 	 	 3814  	 100   	95.50 	1.00 	Jeffrey An, Michael An
 10   Defen 2.2 x64 	  	 3794  	 100   	95.00 	0.00 	Can Catin, Dogac Eidenk
 11   Nemorino 5.00 x64 	 3794  	 100   	95.00 	0.00 	Christian Gunther
 12   Andscacs 0.95 x64 	 3777  	 100   	94.50 	1.00 	Daniel Jose Queralto
 13   Rubi 1.6 x64 	 	 3732  	 100   	93.00 	0.00 	Andreas Matthias
 14   Demolito 2019-11-03 x64 	 3732  	 100   	93.00 	0.00 	Lucas Braesch
 15   Hannibal 1.7 x64 	 	 3732  	 100   	93.00 	0.00 	Sam Hamilton, Edsel Apostol
 16   Nirvana 2.4 x64 	 	 3684  	 100   	91.00 	0.00 	Thomas Kolarik
 17   Senpai 2.0 x64 	 	 3663  	 100   	90.00 	0.00 	Fabien Letouzey
 18   Texel 1.07 x64 	 	 3627  	 100   	88.00 	0.00 	Peter Osterlund
 19   ICE 4.0.853 x64 	 	 3604  	 100   	86.50 	1.00 	Thomas Petzke
 20   Arasan 21.4 x64 		 3516  	 100   	79.50 	1.00 	Jon Dart
 21   Vajolet 2.2.8.0 x64 	 3500  	 100   	78.00 	0.00 	Marco Belli
 22   Ethereal 11.75 x64 	 3278  	 2200    6.59 	0.60 	Andrew Grant

I always suspected that chessowl was pure garbage. Now I know why…

A new entry for my foe list

[Ovyron]

Why not 64 concurrency or 128 concurrency? I mean, if it's going to be about the engine that steals the most resources from others, you can do 256 concurrency and play 800 games instead of 100. Go all the way "saving resources"!

[Raphexon]

Why not 64 concurrency or 128 concurrency? I mean, if it's going to be about the engine that steals the most resources from others, you can do 256 concurrency and play 800 games instead of 100. Go all the way "saving resources"!

Small time CC player.

I added a perft generator to my PGN input, concurrency at 10^43. 2 more seconds and I've solved chess.

[Dann Corbit]

Why not 64 concurrency or 128 concurrency? I mean, if it's going to be about the engine that steals the most resources from others, you can do 256 concurrency and play 800 games instead of 100. Go all the way "saving resources"!

A simple experiment will answer your question.
Find an engine that lets you start more threads than are available on your system.

Now analyze the starting position on a quiet machine with the maximum thread count for your native hardware. Look at the NPS and depth reached after one minute.
Now double the thread count and repeat.
Now double the thread count and repeat.
You will notice that the search is not the same speed but it is severely hampered.
And it gets much worse with each step.

[Raphexon]

Why not 64 concurrency or 128 concurrency? I mean, if it's going to be about the engine that steals the most resources from others, you can do 256 concurrency and play 800 games instead of 100. Go all the way "saving resources"!

A simple experiment will answer your question.
Find an engine that lets you start more threads than are available on your system.

Now analyze the starting position on a quiet machine with the maximum thread count for your native hardware. Look at the NPS and depth reached after one minute.
Now double the thread count and repeat.
Now double the thread count and repeat.
You will notice that the search is not the same speed but it is severely hampered.
And it gets much worse with each step.

I actually just tried to run Stockfish with 512 threads on my PC.
Halted my whole system and then it crashed