On-line engine blitz tourney June

[Joost Buijs]

The monthly on-line engine blitz tourney for June will take place on:

Saturday June 4, 2:00 PM EST (Boston time), 20:00 CET (Amsterdam time)

To connect:

winboard -zp -ics -icshost nightmare-chess.nl -icshelper timeseal -fcp ENGINE.exe -fd ENGINEFOLDER -autoKibitz -keepAlive 30

(for UCI engines, add -fUCI)

People without account can submit a request for an account to admin@nightmare-chess.nl

Our tournament manager will be: Eendje (Lisebeth at ICC)

The format will be: 9 rounds Swiss 5/1

Lisebeth likes to start the tournament on top of the hour, please try to be on-line 15 minutes prior to tournament start, late joins are not supported.

[Krzysztof Grzelak]

Sorry to ask if this tournament can be watched live.

[Joost Buijs]

Yes, this is possible.

You just need a FICS GUI and login as a guest. For instance with Thief you can watch many games at once.

https://sourceforge.net/projects/thief/

[Joost Buijs]

The tourney finished without problems, some regular players were missing, probably due to pentecost.

This time Pontifex (Stockfish 15 on an AMD 5950X) won the tournament. I wonder when Daniel Infuer (if that is his real name) starts competing with something written by himself.

Belfish, ChessSystemTalX and TPinefox (again Daniel Infuer, I assume with LC0) ended second with 6 points.

NightmareX, Seer (run by Ray Banks), Goldbar and Marvin got 5.5 points.

Zappa (the worldchampion of 2005, also run by Ray Banks) still managed to get 4.5 points.

Code: Select all

:Tourney Players: Round 9 of 9 
:
:     Name              Rating Score Perfrm Upset  Results 
:     ----------------- ------ ----- ------ ------ ------- 
:  1 +pontifex          [2917]  7.0  [2992] [  45] +13b +10w +03w =06b =02w +08b =05w +07w =04b 
:  2 +BelFish           [3008]  6.0  [2866] [   0] +12w =08w =04b +10w =01b =03b =06w +11w =05b 
:  3 +ChessSystemTalX   [2889]  6.0  [2859] [  75] +16w +09w -01b +11w +07b =02w =04b =05b =06w 
:  4 +TPinefox          [2500]  6.0  [2914] [1344] =05b +12w =02w =08b +11w =06b =03w +13b =01w 
:  5 +NightmareX        [2909]  5.5  [2820] [  53] =04w =07b =08b +18w =06w +09b =01b =03w =02w 
:  6 +SeerRB            [2902]  5.5  [2820] [  63] +15w +11w =07b =01w =05b =04w =02b =10w =03b 
:  7 +Goldbar           [2872]  5.5  [2786] [  33] +14b =05w =06w +13b -03w =12b +08w -01b +10w 
:  8 +Marvin            [2747]  5.5  [2731] [ 211] +18w =02b =05w =04w +10b -01w -07b +12b +14w
:  9 +WaDuuttie         [2574]  5.0  [2363] [   0] +20w -03b -13w +17b +15b -05w -11b +19w +18w 
: 10 +atomNightmare     [2590]  4.5  [2640] [ 156] +17w -01b +15w -02b -08w +16b +12w =06b -07b 
: 11 +atomGoldbar       [2576]  4.5  [2496] [   0] +19w -06b +16w -03b -04b +17w +09w -02b =15w 
: 12 -TheBaron          [2569]  4.5  [2469] [ 151] -02b -04b +14w +16b +13w =07w -10b -08w +20b 
: 13 +ZappaRB           [2515]  4.5  [2359] [  59] -01w =17b +09b -07w -12b +15w +20b -04w +19b 
: 14 +atomFloyd         [2238]  4.5  [2271] [  35] -07w =18b -12b -15w +20b +19w +17b +16w -08b 
: 15 +Myrddin           [2323]  4.0  [2277] [ 126] -06b +19w -10b +14b -09w -13b =18w +20w =11b 
: 16 +rpiArminius       [2273]  4.0  [2235] [   0] -03b +20w -11b -12w +18b -10w +19b -14b +17w
: 17 +feligres          [2153]  3.5  [2163] [ 206] -10b =13w +20b -09w +19w -11b -14w +18b -16b 
: 18 +CTdccc            [2178]  3.0  [2149] [ 102] -08b =14w +19b -05b -16w +20w =15b -17w -09b 
: 19 +atomSargon        [1844]  1.0  [1945] [   0] -11b -15b -18w +20w -17b -14b -16w -09b -13w 
: 20 +POS               [1483]  0.0  [1896] [   0] -09b -16b -17w -19b -14w -18b -13w -15b -12w 
:
:     Average Rating    2503.0 

[dangi12012]

The tourney finished without problems, some regular players were missing, probably due to pentecost.

This time Pontifex (Stockfish 15 on an AMD 5950X) won the tournament.

Yes - the enemy hardware pool was weaker than in previous iterations. For me - optimizing the hardware itself is also fun - and the participating engine is transparent. I would win against 2970X and 2990X SF15 with my current system.

Like I said earlier - if you were to extend FICS - in addition to finger - there should be the possiblility to query engine IDs and participating hardware automatically by the tournament admin.

That should be included in the results - together with the standard deviation of a 9 round swiss tournament N player result to obtain a sense of a true ranking. (plus engine authors and engine users in seperate buckets)
It makes a difference and you can read this on page 11 of this document: https://repository.upenn.edu/cgi/viewco ... h_scholars

For now its unclear for me how much of the ranking results are noise.

[JVMerlino]

TPinefox is clearly an anagram of pontifex and so most probably run by the same person. There were no finger notes for the engine and no response from the account holder to the question of what engine the account was running. Very disappointing.

[dangi12012]

I wonder when Daniel Infuer (if that is his real name) starts competing with something written by himself.

Might be the wrong thread here but:
If comparing of all possible tree search algorithms takes as long as comparing all sliding algos then it will be mid of this decade.
Its just this way when you explore every avenue and every rabbit hole to find faster code. For the sliding piece lookup it was worth it with the last commit being over 120Billion Queen Moves/s.

There is too much reliance on AB with more or less the same list of pruning techniques.
It is for me like this: Sliding piece Lookup -> Full Movegenerator -> Full Treewalking -> Full Engine

Also looking not to far ahead - zen4 will support BFloat16 (intel has similar plans) which means that cpu and gpu will have the same small floating point op to work with.
For me that hints towards a many times faster version of what NNUE is currently doing. And since much cpu time is spent there one can assume that bfloat16 will play a big role for computerchess in the near future.

Also to add here Not a single chess engine has used tensor cores with wmma::experimental::precision::b1 yet - that can do 8x8x128 matrix multiplication with 2000 TeraOps/s. 8x8 being a perfect alignment with what we do here.
Sudoku programmers dont have it so nice.

[Joost Buijs]

I wonder when Daniel Infuer (if that is his real name) starts competing with something written by himself.

Might be the wrong thread here but:
If comparing of all possible tree search algorithms takes as long as comparing all sliding algos then it will be mid of this decade.
Its just this way when you explore every avenue and every rabbit hole to find faster code. For the sliding piece lookup it was worth it with the last commit being over 120Billion Queen Moves/s.

There is too much reliance on AB with more or less the same list of pruning techniques.
It is for me like this: Sliding piece Lookup -> Full Movegenerator -> Full Treewalking -> Full Engine

Also looking not to far ahead - zen4 will support BFloat16 (intel has similar plans) which means that cpu and gpu will have the same small floating point op to work with.
For me that hints towards a many times faster version of what NNUE is currently doing. And since much cpu time is spent there one can assume that bfloat16 will play a big role for computerchess in the near future.

Also to add here Not a single chess engine has used tensor cores with wmma::experimental::precision::b1 yet - that can do 8x8x128 matrix multiplication with 2000 TeraOps/s. 8x8 being a perfect alignment with what we do here.
Sudoku programmers dont have it so nice.

If you (like you say) don't have an engine ready yet, then why do you enter a tournament with Stockfish and LC0 that is clearly meant for programmers testing their engines?

You sound to me like a 'wise guy', there has been 60 years of development in computer chess, I don't think that there are many things you could improve upon. NNUE is a nice development, only made possible by the speed of current hardware, in the past NN's were just to slow to be useful for computer chess.

BFloat16 will make life somewhat easier, but I don't think that it will be much faster. Stockfish already uses 8 bit quantization and the real bottleneck is memory bandwidth. Intel 'Sapphire Rapids' will get what they call AMX, a matrix multiplication unit for BF16 and INT8, this could help somewhat, but don't expect miracles of it.

The difficult part (training a network that performs well) remains. In case of reinforcement having more speed could help because you can play more games in the same time span, but it won't make a world of difference.

[Joost Buijs]

TPinefox is clearly an anagram of pontifex and so most probably run by the same person. There were no finger notes for the engine and no response from the account holder to the question of what engine the account was running. Very disappointing.

You are right, it is the same person. In one of the last mails I got from him he told me that he would enter the tourney with a weak engine that is developed by himself, instead of doing so he entered the tourney with Stockfish 15 and LC0. Very disappointing indeed.

[RubiChess]

Let me just add the snippet I took from chat in the blitz tourney 11/2021 (the last one I took part) so that everyone may get an idea of this "personality".

Code: Select all

[pontifex]  Why am i not first. Impossible - 32 cores and SF 14.1
[pontifex]  ah sorry my game wasnt over yet. I am first now
[pontifex]  I am on top now - where I belong. This is much better :)
[pontifex]  Im that guy: https://www.codeproject.com/Articles/5313417/Worlds-Fastest-Bitboard-Chess-Movegenerator
[pontifex]  Guys: Who is at place 1 if you type in "tell mamer who 1" ?
[pontifex]  SAY MY NAME
[pontifex]  Guys who is first?
[pontifex]  ??????
+censor pontifex

Regards, Andreas