Page 1 of 1

CCRL 40/40, 40/4 and FRC lists updated (19th October 2019)

Posted: Sun Oct 20, 2019 8:05 am
by Graham Banks
The latest CCRL Rating Lists and Statistics are available for viewing from the following links:
http://ccrl.chessdom.com/ccrl/4040 (40/40)
http://ccrl.chessdom.com/ccrl/404 (40/4)
http://ccrl.chessdom.com/ccrl/404FRC (FRC 40/4)

Our 40 moves in 40 minutes repeating and 40 moves in 4 minutes repeating are both adjusted to the AMD64 X2 4600+ (2.4GHz).
However, our testers use different hardware ranging from the Q8200 through to the most recent i7's and octals. We benchmark our hardware for consistency of results.
These time controls are roughly equivalent to 40 moves in 15 minutes repeating or 40 moves in 1.5 minutes repeating on modern computers.

Please note that the three lists are often updated separately to each other. The FRC list is only updated when a new engine or engine version is being/has been tested.
Also, please note that the 40/40 and 40/4 lists are constructed from totally different databases and not comparable to each other.

Thanks to all of our currently active testers (Graham, Ray, Tirsa, Gabor, Sergio and Volodymyr). Thanks also to Chessdom.com for hosting our website.

40/40 testing this week that I'm aware of will include (with live broadcast port for TLCV noted where applicable):

76th Amateur Series Division 7 Tournament (finishing Thursday 16002)
76th Amateur Series Division 8 Tournament (continuing 16053)
76th Amateur Series Division 9 Tournament (continuing 16001)
New Tournament (starting Thursday 16002)
Topple 0.7.3 64-bit Gauntlet (continuing 16063)
Micah 1.0 64-bit Gauntlet (continuing 16065)
MORA 1.0.0 Gauntlet (continuing 16066)
RofChade 2.202 64-bit Gauntlet (finishing Tuesday 16091)
iCE 4.0 64-bit Gauntlet (starting Tuesday 16091)
Wasp 3.75 64-bit Gauntlet (continuing 16092)
Komodo 13.1 64-bit Gauntlet (finishing Thursday 16093)
Igel 2.0.0 64-bit Gauntlet (starting Thursday 16093)
Googleplex StarThinker 1.6 64-bit Gauntlet (continuing 16094)
Komodo 13.1 64-bit 4CPU Gauntlet (continuing 16083)
RofChade 2.202 64-bit 4CPU Gauntlet (continuing 16084)
Various 4CPU and 1CPU Gauntlets (being run by Volodymyr)
Various 4CPU and 1CPU Gauntlets (being run by Tirsa - on hold)
Various 1CPU Tournaments (being run by Sergio)

40/4 testing since the last update report has included (thanks to Gabor and Sergio):

Winter 0.7 64-bit
ChessBrainVB 3.74
Igel 2.0.0 64-bit
Topple 0.7.2 64-bit
Counter 3.3 64-bit
Counter 3.3 64-bit
DanaSah 7.9
iCE 4.0 64-bit
Sjaak II 1.4.1 64-bit
Raven 0.60 64-bit
RubiChess 1.5 64-bit 4CPU Gauntlet
Demolito 2019-10-08 64-bit
RubiChess 1.6 64-bit 4CPU Gauntlet
Pedone 1.9 64-bit 4CPU
Pirarucu 2.9.5 64-bit 4CPU
Swiss Tournaments (Sergio)

Re: CCRL 40/40, 40/4 and FRC lists updated (19th October 2019)

Posted: Sun Oct 20, 2019 11:59 pm
by Albert Silver
I'm a bit intrigued by the subpar results of the NNs on the 2080 in the 40/4 lists. I know that Leela would absolutely crush the field of 4-core and 8-core participants on a full 2080, so I am assuming this is the result of some calibration process to cripple it in favor of balance. Could you share how it is set up, and what the time control actually played is?

Re: CCRL 40/40, 40/4 and FRC lists updated (19th October 2019)

Posted: Mon Oct 21, 2019 6:05 am
by Graham Banks
Hi Albert,

aside from the number of cores/threads, ponder off, hash size and generic opening books, we test engines with their default settings.

Graham.

Re: CCRL 40/40, 40/4 and FRC lists updated (19th October 2019)

Posted: Mon Oct 21, 2019 4:43 pm
by Albert Silver
Graham Banks wrote: Mon Oct 21, 2019 6:05 am Hi Albert,

aside from the number of cores/threads, ponder off, hash size and generic opening books, we test engines with their default settings.

Graham.
ok cool. I guess I was just expecting a much stronger performance by Leela on a 2080.

Re: CCRL 40/40, 40/4 and FRC lists updated (19th October 2019)

Posted: Mon Oct 21, 2019 5:43 pm
by Jouni
Leela has great results e.g. 94.5 − 65.5 vs Stockfish 10! Not enough?

Re: CCRL 40/40, 40/4 and FRC lists updated (19th October 2019)

Posted: Mon Oct 21, 2019 6:39 pm
by Nordlandia
Graham: cudnn-fp16 give significant better speed than standard cudnn (cudnn is the default) while cudnn-fp16 is clearly better. Why not use the best option?

Leela benefits from Backend cudnn-fp16. No reason to use the worse option just because it's the default :|

If RTX 2080 is used.

Re: CCRL 40/40, 40/4 and FRC lists updated (19th October 2019)

Posted: Mon Oct 21, 2019 7:21 pm
by Modern Times
cudnn-fp16 was used by the tester.

Re: CCRL 40/40, 40/4 and FRC lists updated (19th October 2019)

Posted: Mon Oct 21, 2019 10:42 pm
by Albert Silver
Jouni wrote: Mon Oct 21, 2019 5:43 pm Leela has great results e.g. 94.5 − 65.5 vs Stockfish 10! Not enough?
No, a 2080 with fp16 enabled should beat 4-core SF10 by more than that IMHO.

Re: CCRL 40/40, 40/4 and FRC lists updated (19th October 2019)

Posted: Mon Oct 21, 2019 11:16 pm
by Modern Times
Probably just the blitz time control. Lc0 needs longer time controls to really come alive in my experience.

On the other hand, Stockfish 2019-10-09 on a modest 8CPU machine is a beast at blitz.