Promising engines 3000+ ELO not Stockfish derivatives

[Stephen Ham]

Thanks for posting, Connor. I agree with you.

TalkChess is a diverse mix of decent people and jerks. Thanks to the Moderators for limiting jerk behavior.

Back to the thread's subject, I too fail to see the point of Alex's "blitz" tournament. These engines and others have been, and are still being, tested by several rating agencies. And such testing is performed at longer TCs than "blitz", generating superior evaluations of engine performance.

So if Alex wants to conduct his "blitz" tournament, then I support his enthusiasm for our shared passion. Hurrah! But like you, Connor, I think it's pointless to reinvent the wheel when testing agencies already do it better.

All the best,
-Steve-

[alex67a]

Winter 0.95 NNUE ?

[matejst]

Winter 0.95 is not NNUE. Gabor's compile was huge, so one could think it is. Jonathan uses his own NN implementation.

[AlexChess]

Koivisto 6.15 by Luecx (Finland) 3400+ ELO
https://github.com/Luecx/Koivisto

Koivisto 6.15 SuperBlitz PGN Games https://onedrive.live.com/?cid=B0F79706 ... 69&o=OneUp

New build Koivisto 6.1.6 compiled by IPMAN!!!
much faster and stronger on my tests:
https://mega.nz/file/bIpAlKyA#3Cs3agQdj ... LJy5ZNI6to

(+-3500 ELO and very close to Igel 3.0.5 NNUE!)

Thanks to Lueck that helped me to have it working on my system

[AlexChess]

Winter 0.95 is not NNUE. Gabor's compile was huge, so one could think it is. Jonathan uses his own NN implementation.

Thank you for the info!

Alex

[AlexChess]

Nobody cares.

You know, before i started visiting this forum, i imagined chess programmers as being smart and polite people, that enjoy their hobby and want to share it with others. But damn, i couldn't have been more wrong. There's more drama and toxicity here than i could've ever imagined. It's truly astounding.

+1
Alex is a nice guy who loves computer chess, and has created this topic with great detail and care.
Thx for speaking up and expressing that which many readers of this forum are scared to do...for fear of personal attack from the many haters.

Talkchess seems Afghanistan, not a forum for chess engines enthusiasts and programmers ...Also moving to my own topics is of no help. But there are very interesting and kind people, too (some discovered thanks to the Vintage chess computers post forum3/viewtopic.php?f=2&t=78150 ) and on others similar forums

(So I will NEVER leave my membership here... )

[AlexChess]

Nobody cares.

You know, before i started visiting this forum, i imagined chess programmers as being smart and polite people, that enjoy their hobby and want to share it with others. But damn, i couldn't have been more wrong. There's more drama and toxicity here than i could've ever imagined. It's truly astounding.

+1
I admire Luca!

[matejst]

New build Koivisto 6.1.6 compiled by IPMAN!!!
much faster and stronger on my tests:
https://mega.nz/file/bIpAlKyA#3Cs3agQdj ... LJy5ZNI6to

(+-3500 ELO and very close to Igel 3.0.5 NNUE!)

Thanks to Lueck that helped me to have it working on my system

Thanks for the compiles, Alex.

[AlexChess]

Igel 3.0.5 NNUE by Volodymyr Shcerbyna 3480 ELO (Ukrine)
https://github.com/vshcherbyna/igel/releases/tag/3.0.5

My friend Voffka kindly released also a build for Windows 11 ARM64 on request

This release brings significant improvements in search, time management and evaluation. I accidentally found a strong patch in search which brought around 12 elo in LTC in one shot. Overall strength increase due to search/time management improvements is around 20 elo, strength increase due to new network is 20 elo more. There are also some improvements in SMP mode that were quite good in short time controls: 15 elo, but hard to test on long time control as it will require a lot of computational power.

The new network ign-1-d593efbd is trained from scratch using nnue-pytorch on 12 billions of d8 positions using data from Igel 2.6.0 (HCE), Igel 2.9.0 (NNUE) and Igel 3.0.0 (NNUE) with a validation set of 10 million of d16 data generated using previous network ign-1-139b702b.

Enormous amount of computational power has been used to test this build on OpenBench, I would like to thank Andrew Grant and Bojun Guo (noobpwnftw) for supporting Igel project!

I would like to also thank authors of nnue-pytorch project for helping to make this release possible and more or less 'hassle free' training experience!

What's new:

- Train a new network using Igel 2.6.0 (HCE), Igel 2.9.0 (NNUE) and Igel 3.0.0 (NNUE) data: ign-1-d593efbd
- Better time management in sd time control
- Skip tt cutoff for null moves in search
- Skip null move prunning on tt hit
- Vote for best thread in SMP mode
- Less aggressive time use in non repeated tc

[AlexChess]

Minic 3.1.3 NNUE by Vivien Clauzon (France) 3232 ELO

https://github.com/tryingsomestuff/Mini ... s/tag/3.13

This release is fixing a very important issue with non-AVX2 build of Minic.
The expected strength increase for non-AVX2 release is major.

Code style
For a year and a half Minic was (mainly) a one-file-code with very dense lines. This is of course very wrong in terms of software design... So why is it so? First reason is that Minic was first developped as a week-end project (in mid-october 2018), the quick-and-dirty way, and since then I was having fun going on this way ; being a "small" code size engine was part of the stakes in developing it. Second reason is that it helps the compilers optimize the whole code (at least that is what I though...)

Until version 2 of Minic, some optional features such as Texel tuning, perft, tests, uci support, book generation ... were available in the Add-Ons directory ; they are now fused with the source code.

Nowadays, since the release of version "2" Minic is written in a more classic c++ style, although very dense lines are still present and recall Minic past compacity.