Lc0 41873 Ratings Run 2080ti.

[mwyoung]

Lc0 41873 ratings run.

Hardware 2950x, 2080ti
6 move opening book, optimal book setting.
4 Gb ht max, less for the old test engines.
Default setting, max cpu setting up to 32 threads. Lc0 set to 2 threads.

Current estimated CCRL gauged Elo 3600 to 3650 with 14% of the run completed.

Lc0 41837 Ratings Run..jpg

[mwyoung]

Lc0 41837 Ratings Run 23.5% Complete.jpg

[jorose]

Looking good! Especially the results vs Ethereal and Fizbo look impressive.

Might I enquire about the interesting choice of time control? Not a point of critique, just haven't seen something like it before.

[mwyoung]

Looking good! Especially the results vs Ethereal and Fizbo look impressive.

Might I enquire about the interesting choice of time control? Not a point of critique, just haven't seen something like it before.

It is a time control format that all engines understand well. 2m in 40 is also fast, but still strong on this hardware.

[mwyoung]

After 33% of this ratings run completed to 1000 games. I will say if your are not using Lc0 on good hardware your are missing out. This ratings run is just simple incredible. I am testing Lc0 against some of the best engines available. And after 333 game it's results are crushing. And I am testing Lc0 against using a 16 core AMD 2950x at 4.25 Ghz. Set to the AB engines maximum default strength. And it is crushing all opposition, and this NN is still improving.

3 losses after 333 game. Are you kidding me...


Below is CCRL testing conditions, judge for yourself.

CCRL 40/4
40/4 is the our fast "blitz" time control.

Our members are free to choose any engines they like to test, as long as the testing is done under conditions stated below.

CCRL Testing Conditions
Time Control: Equivalent to 40 moves in N minutes on AMD X2 4600+ at 2.4GHz. We use Crafty 19.17 BH as a benchmark to determine the equivalent time control for particular machine.

We use repeating time control. It means that, say, in 40/40 the engines have 40 minutes for the first 40 moves. Then they get another 40 minutes for the next 40 moves, and so on.

Endgame tablebases: 4 or 5 piece tablebases.

Pondering: OFF.

Tournament format: Any format of tester's choice: Match, Round-robin, Gauntlet, Swiss, etc.

Hash size: Should be set to the same value of either 128 or 256 MB for all engines in a match or tourney. There are two exeptions: 1) Engines using 2 CPUs should have double hash size, compared to single-CPU engines in the same tourney. 4-CPU engines should have 4 times amount of hash. 2) Smaller hash size can be used if an engine has problems with particular hash size, or if it does not allow to configure hash size.

EGTB hash: 32 MB.

Tournament Interface: Any. Examples: Winboard, Arena, Shredder, Chessbase, Chess Partner.

Opening book: Any generic. Examples: remis.ctg, draw.ctg, 5moves.ctg, perfect.ctg etc. Book line length has to be limited to 12 moves per side maximum. The same book should be used for all engines in the same match or tournament.

Engines with their own books should have them disabled (deleted or switched off in parameters). Engines which can't disable their own book can't participate in CCRL testing.

Book learning: Off for all engines.

Position learning: Off for all engines. If learning files exist they must be set to read-only. Any learning during the games played in the rating process is not permitted.

Created in 2005-2013 by CCRL team
Last games added on April 11, 2019

Lc0 41837 Ratings Run 33% Complete.jpg

[Werewolf]

what was the idea of testing the old Schroder engines?

[mwyoung]

what was the idea of testing the old Schroder engines?

Who is testing the old Schroder engines....

This is a ratings run to establish a rating. I have no idea how lc0 will perform. I need a wide range of player strength. Unless now the bottom two programs can at least draw a game. No rating will be established using those game.

The single thread fritz 13 so far is the lowest rated program to draw at least 1 game. But we still have more games to play.

[mwyoung]

Preliminary rating at 51% of the ratings run completed.
Note: Ratings are relative to the pool of players. The ratings number is unimportant.
What is important is the difference in rating, and ranking.

1 Lc0 v0.21.1 3618 2080ti 2 threads
2 Stockfish 090419 64 POPCNT 3600 2950x 32 threads
3 Fire 7.1 x64 popcnt 3505 2950x 32 threads
4 Ethereal 11.25 (POPCNT) 3427 2950x 32 threads
5 Laser 1.7 3413 2950x 32 threads
6 Fritz 16 3370 2950x 32 threads
7 Fizbo 2 3323 2950x 32 threads
8 Fritz 13 SE 3084 2950x 1 thread max threads for this engine
9 Houdini 1.5a x64 3084 2950x 8 threads max threads for this engine

Lc0 41837 Ratings Run 51% Complete.jpg

[supersharp77]

Preliminary rating at 51% of the ratings run completed.
Note: Ratings are relative to the pool of players. The ratings number is unimportant.
What is important is the difference in rating, and ranking.

1 Lc0 v0.21.1 3618 2080ti 2 threads
2 Stockfish 090419 64 POPCNT 3600 2950x 32 threads
3 Fire 7.1 x64 popcnt 3505 2950x 32 threads
4 Ethereal 11.25 (POPCNT) 3427 2950x 32 threads
5 Laser 1.7 3413 2950x 32 threads
6 Fritz 16 3370 2950x 32 threads
7 Fizbo 2 3323 2950x 32 threads
8 Fritz 13 SE 3084 2950x 1 thread max threads for this engine
9 Houdini 1.5a x64 3084 2950x 8 threads max threads for this engine

Lc0 41837 Ratings Run 51% Complete.jpg

Why is Mephisto and Rebel 2000 in that tally? They have 0% results You need engines in a pool that strong that at least can post some results....not 0/52...Thx AR

[mwyoung]

Preliminary rating at 51% of the ratings run completed.
Note: Ratings are relative to the pool of players. The ratings number is unimportant.
What is important is the difference in rating, and ranking.

1 Lc0 v0.21.1 3618 2080ti 2 threads
2 Stockfish 090419 64 POPCNT 3600 2950x 32 threads
3 Fire 7.1 x64 popcnt 3505 2950x 32 threads
4 Ethereal 11.25 (POPCNT) 3427 2950x 32 threads
5 Laser 1.7 3413 2950x 32 threads
6 Fritz 16 3370 2950x 32 threads
7 Fizbo 2 3323 2950x 32 threads
8 Fritz 13 SE 3084 2950x 1 thread max threads for this engine
9 Houdini 1.5a x64 3084 2950x 8 threads max threads for this engine

Lc0 41837 Ratings Run 51% Complete.jpg

Why is Mephisto and Rebel 2000 in that tally? They have 0% results You need engines in a pool that strong that at least can post some results....not 0/52...Thx AR

How do you know they will not get a tally....at 51% of the run.
Poor Fritz 13 on 1 thread did not get a tally either for a long time.
It is not important if they do not score, it is only important if they can score.
I covered the spectrum of possibilities.

Unlike some here. I am not a psychic

"You need engines in a pool that strong that at least can post some results....not 0/52...Thx AR "
In case you missed them here are 8 programs that scored some results.

2 Stockfish 090419 64 POPCNT 3600 2950x 32 threads
3 Fire 7.1 x64 popcnt 3505 2950x 32 threads
4 Ethereal 11.25 (POPCNT) 3427 2950x 32 threads
5 Laser 1.7 3413 2950x 32 threads
6 Fritz 16 3370 2950x 32 threads
7 Fizbo 2 3323 2950x 32 threads
8 Fritz 13 SE 3084 2950x 1 thread max threads for this engine
9 Houdini 1.5a x64 3084 2950x 8 threads max threads for this engine