The TCEC Season 9 Superfinal is a 100-game affair, and after 50 games an early iteration of Stockfish 8 is leading handily against Houdini 5, 28-22, with nine wins against three losses. Every game pair has the engines taking opposite sides of the same preselected opening line, which makes Stockfish's lead even more impressive. Two of Houdini's three wins came in openings where Stockfish won its white game as well. So barring some really big improvements coming down the pike from Houdini or Komodo, it looks like the free engine, Stockfish, is also the best one.
Entries in computer chess (57)
The superfinal of the 9th seasons of the Top Chess Engine Competition (or TCEC, which was also the initialism for its original name, the Thoresen Chess Engines Competition) starts one hour after the beginning of the Carlsen-Karjakin match. It will run a wee bit longer though - it's a best-of-100 contest between the latest iterations of Stockfish and Houdini for computer chess supremacy...at least until season 10 starts sometime next year.
The website for all the action is, as usual, this one.
Also, as some of you know the TCEC crew ran a rapid event with all the season 9 engines after the penultimate stage fo the main event, and the big three - Houdini, Komodo, and Stockfish all went undefeated and went 1-2-3, with Houdini the surprise winner with an enormous score of 56/62. Komodo took second with 53 points, and Stockfish finished a further half a point behind.
There are only 45 shopping days left until Christmas, so if you're looking for something to stick under the tree you'll want to read a different article. If you're looking for something to stick under your analysis board in ChessBase or another GUI, however, you're in luck, as the Big Three - Houdini, Komodo, and Stockfish, in alphabetical order - have all released new iterations of their engines.
Houdini's is the most notable of them all, as it's the first new release in three years. A development version 30 points weaker than the current release qualified for the TCEC Season 9 Superfinal and won the rapid event immediately following the penultimate stage of the normal TCEC tournament, so it's a seriously strong engine.
Komodo's latest version (10.2), is 22 points stronger than 10.1, which is a notable improvement. Of course with Komodo, the sensible thing to do is to buy a year's subscription, which will generally result in approximately four upgrades over that period.
Finally, Stockfish 8 has just been released. As Stockfish is free and has typically been just as good as the other two engines, it's always a winner. Indeed, unless you're a chess professional of some sort (teacher, master level player) or a serious correspondence player, there's no need to go beyond Stockfish unless you like playing around with different engines. They do tend to have different strengths and weaknesses, so if you're using them for opening preparation or deep analysis of strategically complicated positions you may well get different results from each engine.
I'm not a big fan of the annual ICGA World Computer Chess Championship, as the engines don't have to play on the same hardware. Last year Jonny nipped Komodo; or rather, Jonny on 2400 cores barely beat out Komodo on a measly 48. (Rumor has it that Jonny's programmers used meldonium and EPO, and had a direct line in their rest area that let them consult with Veselin Topalov and Vladimir Kramnik. Next year they might be given the option of replaying any game they lose until it comes out in their favor.*)
This year, despite the organizer's best efforts at ensuring fairness (i.e. changing nothing at all), Komodo won anyway. In a bit of a reverse from last year, Jonny outscored Komodo when it came to their results against the rest of the field, but Komodo won the head-to-head. That resulted in a tie, and then Komodo eventually won the third set of tiebreaks to take the title.
While I've been playing the violin for Komodo, someone should rosin up their bow for Shredder. That program, which is far below Komodo in the computer rating lists, took third, only half a point behind Komodo and Jonny, and did so using only 36 cores. It was a small field, and the other three engines were comparatively weak (or perhaps we should say the engine + hardware combinations were relatively weak). So maybe its near-miss was something of a fluke. Unfortunately, we'll never know from a competition like this.
A further report, games, and a sales pitch for ChessBase's version of Komodo, are here.
* N.B. for the humor-impaired and for those who think Jonny's colossal advantages are fair: the "rumors" in the parenthetical are all satirical fictions.
From what I recall of previous TCEC events, the draw death was limited to most of Gull's games and the Komodo-Stockfish finals, so this year's Stage 2 is quite unusual: eight games, eight draws. (It's still very early, obviously, but unusual all the same.)
Stage 1a of season 9 of the TCEC is history, and Komodo (then version 9.42; the just-released version 10 will takes its place in stage 2) won with a super score of 14/15, giving up just two draws. Its chief rival, Houdini 4, was also undefeated going into their head-to-head battle in the last round, but lost to Komodo (I've posted the game here) finish a point and a half behind, a point ahead of fellow old-timer Rybka 4.1.
All of those programs and more qualified for stage 2, which will include the successful programs from stage 1b. The top program in that part of the draw is Stockfish, which surprisingly didn't manage to win in round 1 - albeit with Black - against a program (Vajolet) rated 301 points below it.
You can follow the live action here.
Even though no one is actually playing chess in the Top Chess Engines Championship (formerly known as the Thoresen Chess Engines Competition), strictly speaking, it is reasonable on another level to view it as the real world chess championship, as the best engines - currently the latest versions of Komodo and Stockfish - spit out moves at what would be a 3300+ rating clip if they came from the minds of mortal men. As such, the games are of interest, even if they are not always as accessible as battles between humans. (Which are themselves sometimes relatively inaccessible, both because of the strength of the strongest humans, and because their ideas are often the product of a collaboration with chess engines.)
The event, which will last several months, starts with a field of 32 engines who will be whittled down, step by step, to a final between Komodo and Stockfish the two survivors of the three preliminary stages. (More details here.) As I've already suggested twice, Komodo and Stockfish are significant favorites to reach the final for the fifth time in the last six seasons, but perhaps Houdini will break up the party. The current version of Houdini is quite old - it came out in late 2013 - but its programmer, Robert Houdart, has promised that a new version will come out at some point during the competition. As TCEC rules allow switches to upgraded versions after each stage (assuming the engine in question has qualified for the next stage), Houdart still has a fair amount of time to make his improvements before the start of Stage 3.
(HT: Howard Sample, for reminding me that the event had started.)
Nearly a year ago I purchased Komodo 9 and a one-year subscription, meaning that whenever a new version came out during that time it could be downloaded for no further cost. I have no complaints about the engine, but their notification policy is less than impressive - there are no notifications. (This despite my requesting to be put on a list, and the representative for the company agreeing to do so!) When a TCEC competition is ongoing it's easier to notice when an upgrade comes out, but nowadays it's easier to miss. Version 9.4 came out March 18, and by accident I discovered that a further mini-upgrade came out March 21 - version 9.42.
So for those of you who might have bought the one-year subscription when version 9 first came out, be sure to download the latest and greatest version - it is stronger than its predecessor, and the year is coming to an end in about 3 weeks.
[N.B. The title should not be taken to imply that the Komodo program is disappearing. As far as I know, the company is in good health and they will continue improving their engine indefinitely.]
The chess engines are at it again. Komodo 9.4 won an odds match against GM Joel Benjamin last week, 2.5-1.5. (HT: Vladimir) Joel Benjamin is a good GM, but not as active as he used to be and not as strong as Hikaru Nakamura, who lost an odds match to Komodo last year, so the odds he received were even greater than those given Nakamura:
- Five moves within the first four ranks.
- Rook for knight (a8 for b1, Wr moved to b1) and move.
- f7 pawn removed and two moves.
- Queen for two bishops.
The computer won the first game and drew the remaining three. Interestingly, Nakamura also did fine in the material odds games, but also lost the "free tempi" game (though he "only" got four moves). The next human sacrifice will be Eugene Perelshteyn in April; hopefully there will be a man-bites-computer story to tell for a change.