Introducing Igel chess engine

[voffka]

Thank you. It is still not working, I sent you a PM with the details.

Sent you another compile, with SSE 3 and SSE 4 instead of SSE 4.1.

[Chessqueen]

Thank you. It is still not working, I sent you a PM with the details.

Sent you another compile, with SSE 3 and SSE 4 instead of SSE 4.1.

Where are the executable files of Igel

[voffka]

Thank you. It is still not working, I sent you a PM with the details.

Sent you another compile, with SSE 3 and SSE 4 instead of SSE 4.1.

Where are the executable files of Igel

The official Igel executable are at https://github.com/vshcherbyna/igel/releases/tag/2.9.0

The binaries I compile for people are custom and best suited for their cpu architecture (in terms of performance), so it is not a general purpose releases, hence I will not publish them on the project page.

[Sylwy]

Igel 2.9.0 at https://github.com/vshcherbyna/igel/releases/tag/2.9.0 (official executable binaries for Windows and IGN net).

After three months of generating data using Igel's own search and evaluation and training several network architectures I am happy to announce a new version of Igel 2.9.0 that uses it's own network trained on Igel search/eval data (around 2.3 billions of depth 8 and 500 millions of depth 12). Earlier versions of this network seen strong participation of Igel in TCEC Season 20 and should bring more than 20 elo of strength at LTC.

What's new:

- Switch to own network based on Igel search and evaluation (ign-0-9b1937cc)
- NNUE dynamic eval scaling

Regression run against Igel 2.8.0 at 60s+0.6:

Code: Select all

ELO   | 24.06 +- 5.30 (95%)
SPRT  | 60.0+0.6s Threads=1 Hash=64MB
LLR   | 0.00 (-2.94, 2.94) [0.00, 0.00]
Games | N: 7376 W: 1908 L: 1398 D: 4070

+85 Elo points (after only 13 games) in my test vs. SlowChess Blitz Classic 2.4 x64 (TC=4'+2").

[voffka]

Your sample size is too small You will need to run around 16k long time control games to get something precise, here is the run from updated regression:

http://chess.grantnet.us/test/9069/

Code: Select all

ELO   | 23.78 +- 3.56 (95%)
SPRT  | 60.0+0.6s Threads=1 Hash=64MB
LLR   | 0.00 (-2.94, 2.94) [0.00, 0.00]
Games | N: 16112 W: 4102 L: 3001 D: 9009

[Sylwy]

1.-Your sample size is too small
2.-:)
3.-You will need to run around 16k long time control games to get something precise,

1.-I know. In fact is just the overture of my test (100 games at TC=4'+2" - 12-15 minutes per game);
2.-
3.- This is a job for CCRL, CEGT & co. My tests are just orientative. Nothing more.

[Jamal Bubker]

Hi Volodymyr !

Thanks for the new release !
Please could you provide a windows binary of Igel 2.9.0 designed for Sandy Bridge or for westemere architecture that supports only SSE4.1, SSE4.2 and POPCNT.
My processor is a 4 cores Xeon E3-1220, 3,1 Ghz.
Thank you
Greetings,
Jamal

[voffka]

Hello Jamal,

Sure, I will send you a customized compile today by EOD via PM.

[voffka]

Igel 2.9.0 optimized for westemere is provided to Jamal over PM.

[Jamal Bubker]

Thank you so much Volodymyr