nTCEC

Some Personal Reflections on nTCEC

nTCEC is simply the highest quality computer chess tournament staged.   It is run on big harware with slow time controls which gives rise to good quality chess.   For a tournament it has an awful lot of games making the results less prone to statistical randomness and therefore more meaningful (sure we would all like 1000s of games to make the results more robust but that is not realistic at this time control).  It is not surprising that nTCEC is often referred to as the de facto world computer championship.   As far as I can see it is a better tournament than any "official" world chess championship has ever been.

I look forward to season 3 greatly but also with a few concerns.

Firstly time controls.   The time limit was speeded up between season 1 and 2.   Some people watching wanted it even faster.    I am sure if it is speeded up again there will still be people wanting it faster still.   Will we eventually end up with a tournament at 20 mins per game because the majority of people find that the most entertaining?   I am sure fast games are the way to get more interest and will gradually win out in any poll for time controls.    However, we would lose what makes nTCEC special and distinguishes it from so many other tournaments.

Popularity breeds.....?   The quality of chat on nTCEC season 1 was overall pretty good.    The quality of chat on nTCEC season 2 was IMHO overall pretty dire with some notable exceptions and by and large I did not participate in it.    I am not sure that anythnig can be done about this.    

Tablebases.   There are two things aobut TBs that concern me re nTCEC.   For engines that use them I would very much like to see 6 man TBs reintroduced.   Its the same theme for me as the time control - quality of play.   I understand that there is a cost issue here as more SSD space would be required, so it will entirely understandable if we have to go without 6-man TBs.   The other thing is the GUI using TBs for adjudication when the engines don't use them.   This seems bizarre as it could give rise to results that the engines would not reach without adjudication.   Perhaps with 5 man TBs only this is less of an issue but it would be more risky with 6 man TBs.  
  
Openings.    This is a difficult topic and it is perhaps impossible to keep everyone happy on this one.    I am not going to write much on this at this point as my jury is out on the new more systematic approach to selecting openings that was used in the Superfinal of season 2 and which is apparently going to be extended into season 3.   Maybe it will be a good way of throwing light on where engines most need to develop or maybe it won't.   Maybe it will give rise to interesting chess or maybe it won't. 

Regardless of these concerns I am sure nTCEC season 3, expected to start late January 2014, will be superb once more.   It looks like it remain the de facto world computer championship for some time :)

nTCEC season 2 summary (now renamed TCEC season 5)

Results

My first post on this blog was a summary of the results of the first nTCEC season.   Since then Season 2 has taken place with Komodo being the winner, beating Stockfish in the final.

Someone else, Martijn Grimme, has done a superb job in documenting the results of season 1 and season 2.   http://chess.martijngrimme.nl/stage.php
There is little point in me duplicating his work.

However, there are a few extra things that don't appear in Martijn's results that are worth recording for posterity's sake.

Conditions

nTCEC season 2 ran from late August to the beginning of December in 2013.

It ran on a Dual Xeon E5-2689 @ 3.3 GHz (i.e 16 cores) with 32 GB off RAM.   This was the same hardware that was used for later stages of nTCEC season 1.

Engines ran with ponder off and were allowed to use up to 8GB of hash and up to 1GB of tablebase hash.

The time limit was 120 minutes + 30 second increment.   (This was appreciably faster than the 150 minutes + 60 seconds allowed in season 1).

5-men Syzygy, 5-men Nalimov, 5-men Gaviota (cp2), 5-men Shredderbases and 5-men Robbobases (Totalbases + Triplebases) were available depending on which each specific engine can use. For some engines specifying more than one type is possible, but here only one was allowed.  They were hosted on a dual Samsung SSD Raid 0 setup.

This varied from season 1 in so far as:
- 6-men Nalimovs were dropped having being introduced  part of the way through nTCEC season 1
- Syzygy's were introduced for the first time having been released in the summer of 2013.

The GUI used TBs to adjudicate games regardless of whether the engines used TBs.

Engine updates were allowed between rounds but not during rounds.  Some authors took this opportunity whilst others did not.

Notes on the leading engines.

Unlike in season 1 of nTCEC, Komodo was multi-core.   It was updated at each stage of the tournament.   Before the start of the tournament Komodo 5 had been publicly released, during the tournament Komodo 6 was publicly released and at the end of the tournament Komodo TCEC was released, this being the version that played in the nTCEC season 2 superfinal.   All these versions of Komodo ran without tablebases.

Stockfish was also updated at each stage of the tournament.   The Stockfish version that started the tournament was stronger than Stockfish 4.   The version that played in the superfinal was released as Stockfish DD.   At the time the tournament was running (and still at the point of writing this) the versions of Stockfish produced by its official developers did not support tablebases.    However, other developers have produced versions of Stockfish that support TBs.   In Stage 3 a version of Stockfish using Syzygy TBs played.   In all other rounds versions of Stockfish produced by the offical team (i.e. using no TBs) played.

Houdini 3 (the publicly released version) played in Stages 1 and 2.    Houdini 9601, a development version, played in Stages 4 and 5.    People were commonly referring to this as "Houdini 4 beta" though as far as I am aware no public statement was made to this effect by its author, Robert Houdart, so we do not know how close to Houdini 4 this version was.    Regardless, it did not reach the superfinal, having come a close third in stage 4.   Houdini 3 ran with Nalimovs and Houdini 9601 with Syzygy TBs.

Don Dailey

Don Dailey and Larry Kaufmann were the original authors of Komodo with Don being the computer guy (who could also play chess!) and Larry being the chess expert.   Don had been fighting leukaemia for some time but his condition worsened dramatically during the tournament and during this period he was admitted to hospital for the final time.  Very sadly he died during the superfinal.

Don regularly participated on nTCEC chat in both season 1 and 2 and was always thoughtful, polite and fun.   He will be missed for that as well as his other contributions to chess and wider life.

During the tournament it was announced that Mark Lefler would be taking over from Don on the Komodo team.

FICGS correspondence chess.

FICGS is the Free Internet Correspondence Games Server.    http://www.ficgs.com/

It allows people to play Go and Poker on-line but is best known as one of the leading on-line correspondence chess servers.  As the name implies it is FREE to use.   It is also a polite and friendly place to play.

You can enter tournaments at different rates of play.   "Standard" is 10 moves in 40 days.   "Rapid" is 30 days for all your moves plus a 1 day increment each time you make your move.   "Blitz" is one hour plus one minute increment per move.   "Lightning" is 20 minutes plus 30 seconds per move.   You enter a waiting list for a tournament and when there are enough players the tournament starts.   You receive an email notification when the tournament starts and each time your opponent moves, though you can also just stay logged in and see your games through your browser any time you like.  Personally I really like the Rapid rate of play but may try the Standard in the near future.

Use of computers to help analyse games is allowed (and indeed encouraged) in most tournaments, though there are some "no engines" tournaments for the purists.

FICGS operates an ELO style rating systems and as your rating progresses you are eligable to enter stronger tournaments.

FICGS runs a world championship which is open to everyone, regardless of strength and including new members.   It is a 5 stage event, so despite being played at the Rapid speed each world championship cycle takes several years.   The current champion is the well known and extremely strong Italian correspondence player Riccio Eros.   

The next FICGS world championship is scheduled to start on 1st March 2014 and the waiting list to enter will be open for a few weeks before that.

More details on FICGS are here http://www.ficgs.com/membership.html
and specificallly its Chess World Championship rules here http://www.ficgs.com/membership.html#chess_wch

Details of previous FICGS championship results are here https://skydrive.live.com/?cid=afb49fe3a5961d46#cid=AFB49FE3A5961D46&id=AFB49FE3A5961D46!231
 

Summary of Conditions and Results for nTCEC Season 1 (now renamed TCEC Season 4)

Summary of Conditions and Results for nTCEC Season 1  (now renamed TCEC Season 4)       

http://www.tcec-chess.net/
nTCEC is a slow time control tournament for chess engines.   Whilst not called the world championship it is clearly the most testing of tournaments in terms of the number of games, quality hardware and slow time control.    Much kudos to Martin Thoresen who runs it.    Season 2 will start later this year.


Introductory key notes

This was a 5 stage tournament culminating in the Superfinal

The hardware changed after Stage 2, from a 4 core machine to a 16 core machine
 
3,4 and 5 man tablebases were available for engines in various formats for Stages 1 and 2.   6 man Nalimovs were added for Stage 3 onwards.
 
Stockfish was updated between stages.   The version in the Superfinal was very similar to Stockfish 3
 
Komodo 4534 ran on a single core and was later released as Komodo CCT
 
Chiron was withdrawn in Stage 3 due to a number of crashes and its scores were removed from that stage.


Superfinal 

150m + 60s per move for the whole game,
running on a dual Intel Xeon E5-2689 @ 3300 MHz.
16 cores used for each engine, ponder off.   
Complete 3,4 and 5 man EGTBs, most important 6 man EGTB.

Houdini 3   25-23   Stockfish (very similar to Stockfish 3)



Stage 4

150m + 60s per move for the whole game, 
running on a dual Intel Xeon E5-2689 @ 3300 MHz.
16 cores used for each engine, ponder off.   
Complete 3,4 and 5 man EGTBs, most important 6 man EGTB.

N Engine            Pts    SB Houdin Stockf Rybka  Komodo

1 Houdini 3        11.0 89.50        100=11 ====== =1==11
2 Stockfish 120413  9.5 81.75 011=00        ==1=== ==1===
3 Rybka 4.1         9.0 79.50 ====== ==0===        ====1=
4 Komodo 4534       6.5 62.75 =0==00 ==0=== ====0=      



Stage 3
150m + 60s per move for the whole game, 
running on a dual Intel Xeon E5-2689 @ 3300 MHz.
16 cores used for each engine, ponder off.   
Complete 3,4 and 5 man EGTBs, most important 6 man EGTB.

N Engine            Pts    SB Ho St Ry Ko Vi Hi Qu Ch

1 Houdini 3        10.0 56.50    == 01 == =1 1= 11 11
2 Stockfish 250313 10.0 54.75 ==    0= == =1 11 11 11
3 Rybka 4.1         9.5 55.50 10 1=    == == == 11 11
4 Komodo 4534       9.5 54.50 == == ==    01 11 1= 11
5 Vitruvius 1.19    7.0 38.00 =0 =0 == 10    0= 1= 11
6 Hiarcs 14         6.0 29.00 0= 00 == 00 1=    == 11
7 Quazar 0.4        4.0 14.25 00 00 00 0= 0= ==    11
8 Chiron 1.5        0.0  0.00 00 00 00 00 00 00 00   
Stage 2

150m + 60s per move for the whole game, 
running on an Intel Core i7 3770k @ 4500 MHz.
Up to 3 cores were used for engines capable of SMP.   
Complete 3,4 and 5 man EGTBs in various formats.
Stage 2a                                               Stage 2b

N Engine           Pts  Ho St Ry Hi Cr Ju Sp Ga        N Engine           Pts  Vi Ko Ch Qu Ha Gu Sh Sc   


1 Houdini 3        9.0     == 01 == 01 11 1= =1        1 Vitruvius 1.19  10.0    == 1= =1 =1 1= =1 1=
2 Stockfish 210213 9.0  ==    == == == 1= 1= 11        2 Komodo 4534      9.5  ==    == == 10 =1 11 11
3 Rybka 4.1        8.5  10 ==    == 1= =1 =1 ==        3 Chiron 1.5       8.0  0= ==    =1 == 01 01 11
4 Hiarcs 14        8.5  == == ==    0= == 11 11        4 Quazar 0.4       7.5  =0 == =0    1= 01 1= =1
5 Critter 1.6a     8.0  10 == 0= 1=    == == 11        5 Hannibal 200213  6.5  =0 01 == 0=    01 == 1=
6 Junior 13.3      5.5  00 0= =0 == ==    =0 11        6 Gull R375        6.5  0= =0 10 10 10    == =1
7 Spike 1.4        5.0  0= 0= =0 00 == =1    01        7 Shredder 12      5.5  =0 00 10 0= == ==    1=
8 Gaviota 0.8602   2.5  =0 00 == 00 00 00 10           8 Scorpio 2.75     2.5  0= 00 00 =0 0= =0 0=   




Stage 1

150m + 60s per move for the whole game, 
running on an Intel Core i7 3770k @ 4500 MHz.
Up to 3 cores were used for engines capable of SMP.   
Complete 3,4 and 5 man EGTBs in various formats.
Engine               Pts    CS Bon  Elo       1       2       3       4       5       6       7

0001 Houdini 3.0     5.5 30.00 0.0 3193 0024:w+ 0005:b= 0007:w= 0004:b+ 0008:w+ 0003:b= 0011:w+
0002 Vitruvius 1.19  5.5 28.00 0.0 3077 0028:b+ 0014:w= 0010:b= 0012:w+ 0006:b+ 0005:w+ 0003:b=
0003 Gull II         5.0 30.00 0.0 3091 0027:b+ 0016:w+ 0006:b= 0009:w+ 0007:b= 0001:w= 0002:w=
0004 Stockfish 2.31  5.0 27.00 0.0 3092 0025:w+ 0021:b+ 0009:b= 0001:w- 0010:b= 0014:w+ 0008:w+
0005 Hiarcs 14.0b    4.5 30.00 0.0 2981 0022:b+ 0001:w= 0011:b+ 0006:w= 0009:b= 0002:b- 0015:w+
0006 Chiron 1.5      4.5 29.00 0.0 2996 0023:w+ 0012:b+ 0003:w= 0005:b= 0002:w- 0015:b= 0017:w+
0007 Komodo 4534     4.5 28.50 0.0 3124 0019:b+ 0010:w= 0001:b= 0021:w+ 0003:w= 0008:b- 0016:w+
0008 Rybka 4.1       4.0 29.50 0.0 3093 0011:b= 0017:w= 0014:b+ 0018:w+ 0001:b- 0007:w+ 0004:b-
0009 Critter 1.6a    4.0 29.00 0.0 3089 0018:w+ 0015:b+ 0004:w= 0003:b- 0005:w= 0016:b= 0012:w=
0010 Shredder 12.0   4.0 27.50 0.0 2958 0032:w+ 0007:b= 0002:w= 0015:b= 0004:w= 0012:b= 0013:w=
0011 Scorpio 2.75    4.0 27.00 0.0 2807 0008:w= 0013:b+ 0005:w- 0029:b+ 0015:w= 0019:b+ 0001:b-
0012 Junior 13.3     4.0 25.00 0.0 2900 0030:b+ 0006:w- 0028:b+ 0002:b- 0024:w+ 0010:w= 0009:b=
0013 Gaviota 0.86b3  4.0 23.50 0.0 2720 0017:b= 0011:w- 0020:b+ 0016:w- 0029:b+ 0021:w+ 0010:b=
0014 Quazar 0.4      3.5 26.50 0.0 2897 0026:w+ 0002:b= 0008:w- 0023:b= 0027:w+ 0004:b- 0018:w=
0015 Hannibal 1.3    3.5 26.00 0.0 2945 0031:b+ 0009:w- 0019:b+ 0010:w= 0011:b= 0006:w= 0005:b-
0016 Spike 1.4       3.5 25.50 0.0 2919 0029:w+ 0003:b- 0023:w= 0013:b+ 0021:b= 0009:w= 0007:b-
0017 Equinox 1.65    3.5 25.50 0.0 3049 0013:w= 0008:b= 0021:w- 0025:w+ 0018:b= 0020:w+ 0006:b-
0018 Texel 1.01      3.5 22.00 0.0 2797 0009:b- 0031:w+ 0022:b+ 0008:b- 0017:w= 0027:w= 0014:b=
0019 Nemo 1.01b      3.5 21.50 0.0 2851 0007:w- 0032:b+ 0015:w- 0028:b+ 0023:w+ 0011:w- 0022:b=
0020 The Baron 3.34b 3.5 19.50 0.0 2581 0021:w- 0025:b= 0013:w- 0032:b+ 0028:w+ 0017:b- 0027:b+
0021 Protector 1.5b2 3.0 27.00 0.0 2907 0020:b+ 0004:w- 0017:b+ 0007:b- 0016:w= 0013:b- 0024:w=
0022 Exchess 6.71b   3.0 23.00 0.0 2680 0005:w- 0024:b+ 0018:w- 0027:b- 0026:w+ 0023:b= 0019:w=
0023 Arasan 15.1     3.0 22.50 0.0 2698 0006:b- 0030:w+ 0016:b= 0014:w= 0019:b- 0022:w= 0025:b=
0024 Minkochess 1.3  3.0 21.50 0.0 2847 0001:b- 0022:w- 0031:b+ 0030:w+ 0012:b- 0025:w= 0021:b=
0025 Crafty 23.5     3.0 21.00 0.0 2763 0004:b- 0020:w= 0029:w= 0017:b- 0032:w+ 0024:b= 0023:w=
0026 Danasah 5.0     3.0 16.50 0.0 2517 0014:b- 0028:w- 0027:b- 0031:w+ 0022:b- 0029:w+ 0030:b+
0027 Dirty 190113    2.5 23.50 0.0 2723 0003:w- 0029:b- 0026:w+ 0022:w+ 0014:b- 0018:b= 0020:w-
0028 Octochess 4741  2.5 22.00 0.0 2678 0002:w- 0026:b+ 0012:w- 0019:w- 0020:b- 0030:w= 0032:b+
0029 Rodent 0.17     2.0 21.50 0.0 2602 0016:b- 0027:w+ 0025:b= 0011:w- 0013:w- 0026:b- 0031:w=
0030 Nebula 2.0b     1.5 18.00 0.0 2421 0012:w- 0023:b- 0032:w= 0024:b- 0031:w= 0028:b= 0026:w-
0031 Redqueen 1.13   1.5 17.50 0.0 2600 0015:w- 0018:b- 0024:w- 0026:b- 0030:b= 0032:w= 0029:b=
0032 Prodeo 1.83c    1.0 19.50 0.0 2607 0010:b- 0019:w- 0030:b= 0020:w- 0025:b- 0031:b= 0028:w- 






www.ficgs.com  is an excellent, friendly and free corrrespondence chess, go and poker site.