Commercial Release of Ethereal 13.00 (NNUE) for AVX2 Systems

[AlexChess]

Most testers (and me too) have reported that Ethereal 13.00 NNUE commercial is 20 ELO stronger than Igel 3.0.5 NNUE,

I dont. In my testings Ethereal 13 nnue is +59 Elo stronger than Igel 3.0.5 nnue. Both engines have played 8000 games. So, the errorbar is +/-6 Elo, only, for both engines.

I think +75 Elo, in my ratinglist, of Ethereal 13 nnue compared to Ethereal 12.75 is a valid result, when mentioning, that A.Grants selfplay progression test was:

ELO | 121.51 +- 5.35 (95%)
CONF | 10.0+0.1s Threads=1 Hash=8MB
Games | N: 10436 W: 4966 L: 1458 D: 4012

So, in my testruns, using longer thinking-time (3'+1'' (but using slower hardware (Notebook, running engines on 20 Hyperthreading-Threads on a 12 core machine)), I would have expected an Elo-gain of around +70 to +80 Elo. And thats exactly, what we see.
The +45 or so Elo-gain of other testers seem pretty strange to me.

https://www.sp-cc.de

Hi!
Could you also test Ethereal 13.0.0 github standard (not NNUE) to see the differences?

Best regards, Alex

[AlexChess]

Stockfish open source philosophy has changed computer chess forever.
Cooperative programming offers today hundreds of Word-Champion-level engines for free.
Indipendent engines cannot reach the performances of Stockfish and derivatives in the near future, because a team of 20 skilled programmers and thousands of testers are absolutely stronger than a very skilled programmer writing code for himself.
I'm very grateful to the Stockfish Team: Tord Romstad, Marco Costalba, Joona Kiiski, Andrea Manzo, Ronald de Man, Michael Byrne and all the others who share their own engine with the community.

Andrew, you should let (also) to setup and develop custom NNUEs on Ethereal 13.00 github, to improve your program and all the others





All the games played 3 mins + 3 secs (no opening book, with 6-man syzygy tablebases, 4 CPUs up to 10 - 15 mN/s on Mac mini M1)


PNG WITH ALL GAMES: https://onedrive.live.com/?cid=B0F79706 ... 69&o=OneUp

[mclane]

What i can say so far is that ethereal with NNUE is stronger then stockfish 13 without NNUE,
This is consistent on all my machines.

[ernest]

According to Stefan Pohl :

Ethereal 13 nnue avx2 : 3475
Stockfish 12 200902 : 3688
Stockfish 11 200118 : 3568

[Cornfed]

What i can say so far is that ethereal with NNUE is stronger then stockfish 13 without NNUE,
This is consistent on all my machines.

Again, comparing apples with oranges one can say lots of things...

[AlexChess]

What i can say so far is that ethereal with NNUE is stronger then stockfish 13 without NNUE,
This is consistent on all my machines.

Again, comparing apples with oranges one can say lots of things...

Why? It's a programmer's choice to add or not NNUE to his engine. I test both. Also, if the author offers his own optimized opening library it would be far to use it, but few engines include it, so I prefer to disable opening library on all engines.

I set for all the partecipants the same number of CPUs and TB Sygyzy 6-man, avoiding loses on time with 3 mins + 3 secs / all and I repeat all crashed games.

If I cannot use an engine (like Ethereal 13.0.0 and Pedone 3.1, *the only crashing with more than 1 CPU between 50+ tested on my hardware*) I extrapolate their rating with separate gauntlets then add the results



Now running a gauntlet Ethereal 13.00 standard (github) against all the others using only 1 CPU, to verify if it is stronger than Ethereal 12.75 NNUE... Stay tuned

[AlexChess]

All the games of my SuperBlitz 38 - 703 games 3 mins + 3 secs / all on Banksiagui 0.45 for Windows 10 ARM64 under Parallels Desktop M1 4 CPUs (2-4 mN/s average)

https://onedrive.live.com/?cid=B0F79706 ... 69&o=OneUp

Ethereal 13.00 standard (github) 1 CPU for all them will follow

[Chessqueen]

According to Stefan Pohl :

Ethereal 13 nnue avx2 : 3475
Stockfish 12 200902 : 3688
Stockfish 11 200118 : 3568

So equal to Stockfish 10

[AlexChess]

Completed my Ethereal 13.00 github gauntlet to verify the differences with the commercial version (not compatible with my hardware )

[AlexChess]

Gauntlet Ethereal 13.00 Github standard (no NNUE) vs all other top engines 3 min + 3 secs 1 CPU (1,2 MN/s average)



ALL GAMES OF THE GAUNTLET: https://onedrive.live.com/?cid=B0F79706 ... 69&o=OneUp

Despite the fact that is the faster engine, is scores only 14/38 points (30th), so Ethereal 12.75 + Stockfish NNUE c4c3... is better (20th) Every engine without NNUE today is obsolete out-of-the-box