STARTED THE NEW 36 GAMES MATCH AGAINST ETHERLITO (ETHEREAL 12.69 NNUE)
ProteusSF-Aureo-240405 running at 1 / 25 of Etherlito speed (130 kN/S vs 2.62 MN/s) 8moves_V3.pgn books for both - Tablebases Syzygy 6 man.
Will Etherlito 12.69 NNUE score better than Ethereal 14.32 github against ProteusSF-Aureo 240405?
AlexChess wrote: ↑Thu Apr 18, 2024 9:21 am
After 36 long time games 25+1, even if Ethereal 14.32 performance is a little worse compared to the beginning of the test, +11 / 36 matches what I was expecting and I still think that is very interesting to study it to learn all the modern chess coding techniques. Probably it is the best open source engine without NNUE, so for didactical purposes and historic archives about the no NNUE era is perfet. I will now compare it with Etherlito (first Grant's trial about NNUE) to verify if, after 2 years of development, classical evaluation has reached the level of the rudimental 2021 NNUE boosted engines.
Uri Blass wrote: ↑Thu Apr 18, 2024 2:08 pm
The result is clearly worse for Ethereal than what I expect and I suspect the opposite use worse engine or inferior hardware.
I do not have Ethereal but here are the results of it in the CCRL rating list against top stockfish versions that they test
The hardware:
2008 Core 2 Duo 2.4 Ghz (ProteusSF-Aureo-240405) 130 kN/s
8 cores computer (Ethereal 12.69 NNUE) 2.6 MN/s (25 times faster than my old PC)
Ethereal 14.32 github is without the NNUE (5.1 MN/s)
Ethereal 12.69 NNUE is 2 years old (2.6 MN/s)
Nobody tests the official free github version of Ethereal, so I'm curious.
Do you know about other top engines not using NNUE?
There are a lot of spectators on Lichess.org, since i broadcast
the games LIVE all day. I will post all the games here, too.
If somebody want to challenge ProteusSF-Godot (running on minimal hardware for fairplay, being based on Stockfish) please leave me a note here. I can suspend current tests with Ethereal and Rubichess, if needed
AlexChess wrote: ↑Thu Apr 18, 2024 9:21 am
After 36 long time games 25+1, even if Ethereal 14.32 performance is a little worse compared to the beginning of the test, +11 / 36 matches what I was expecting and I still think that is very interesting to study it to learn all the modern chess coding techniques. Probably it is the best open source engine without NNUE, so for didactical purposes and historic archives about the no NNUE era is perfet. I will now compare it with Etherlito (first Grant's trial about NNUE) to verify if, after 2 years of development, classical evaluation has reached the level of the rudimental 2021 NNUE boosted engines.
Note: I have assigned 1/2-1/2 to 3 draw games lost for internet disconnection on both sides.
Etherlito is a clear joke, making fun of the expected future shitty clones. Which came true exactly as I predicted. You running it is top irony.
MODERATORS THIS IS IN THE WRONG SUB-FORUM FIX IT
Friendly reminder that stealing is a crime, is wrong, and makes you a thief. "Those who can't do, clone instead" - Eduard ( A real life friend, not this forum's Eduard )
Uri Blass wrote: ↑Thu Apr 18, 2024 2:08 pm
The result is clearly worse for Ethereal than what I expect and I suspect the opposite use worse engine or inferior hardware.
I do not have Ethereal but here are the results of it in the CCRL rating list against top stockfish versions that they test
The hardware:
2008 Core 2 Duo 2.4 Ghz (ProteusSF-Aureo-240405) 130 kN/s
8 cores computer (Ethereal 12.69 NNUE) 2.6 MN/s (25 times faster than my old PC)
Ethereal 14.32 github is without the NNUE (5.1 MN/s)
Ethereal 12.69 NNUE is 2 years old (2.6 MN/s)
Nobody tests the official free github version of Ethereal, so I'm curious.
Do you know about other top engines not using NNUE?
There are a lot of spectators on Lichess.org, since i broadcast
the games LIVE all day. I will post all the games here, too.
All top engines use NNUE or NN even if they are not stockfish(lc0 is using NN and not NNUE)
I read that Weiss is the best engine that does not use NNUE in TCEC when Xiphos 0.6 64-bit 4CPU is the CCRL leader of not stockfish version that do not use NNUE
I think Stockfish11 is the last non developement version of stockfish that does not use NNUE and is clearly stronger than them and also older versions of stockfish are stronger then them:
Stockfish 11 64-bit 4CPU 3548
Stockfish 10 64-bit 4CPU 3528
Stockfish 9 64-bit 4CPU 3490
Stockfish 8 64-bit 4CPU 3438
Stockfish 7 64-bit 4CPU 3386
AndrewGrant wrote: ↑Thu Apr 18, 2024 9:35 pm
Etherlito is a clear joke, making fun of the expected future shitty clones. Which came true exactly as I predicted. You running it is top irony.
Congratulations! It's a very strong engine to be a joke and the code is very easy to understand even for a dummy cloner like me.
It is compatible with all Stockfish 20530 kb NNUEs. I expect a lot of fun for me on modding it!
I will publish here and on BanksiaGui forum (only the contents without spam) my last surprising results.
Uri Blass wrote: ↑Fri Apr 19, 2024 8:18 am
All top engines use NNUE or NN even if they are not stockfish(lc0 is using NN and not NNUE)
I read that Weiss is the best engine that does not use NNUE in TCEC when Xiphos 0.6 64-bit 4CPU is the CCRL leader of not stockfish version that do not use NNUE
I think Stockfish11 is the last non developement version of stockfish that does not use NNUE and is clearly stronger than them and also older versions of stockfish are stronger then them:
Stockfish 11 64-bit 4CPU 3548
Stockfish 10 64-bit 4CPU 3528
Stockfish 9 64-bit 4CPU 3490
Stockfish 8 64-bit 4CPU 3438
Stockfish 7 64-bit 4CPU 3386
Thank you, you are a wonderful source of infos!! I will test also Weiss an Xiphos LIVE on Lichess.org.
Note: If somebody has commercial Ethereal 14.25 NNUE i'd like to challenge it too on Lichess.org
Very good performance of Ethereal 12.69 + SF NNUE, better than latest updated Ethereal 14.32 github without NNUE
I then choose the older one as the base code for my modded version. Starting right nowe the Live games of Ethereal-Proteus-NNUE on Lichess.org against all other no-Stockfish-bots.