After being away from chess for almost 20 years (with a short return between 2007 and 2009), I've now decided to drop some of my many hobby's permanently and start studying chess again. (Go being one of them... There's nothing like a wooden DGT board for Go.) 20 years ago, my over the board rating fluctuated between 1800 and 1835, with an all time high of 1865; with me never opening a book or reviewing a single game. Nowadays, I'll be surprised if my rating even reaches 1600 :p
Being a software engineer/programmer, and having written a (extremely basic) chess program my teens in Borland Pascal, I've also been an avid computer chess fan. I was curious how strong Stockfish would be on levels other than level 20. A test has been run by Pohl4711, on talkchess:
viewtopic.php?t=69731
While it is a good test to make an educated guess as to the place of the SF levels in the CCRL rating list, that user didn't use the CCRL timings, used ORDO for ratings instead of BayesELO, and calibrated to a different list.
I am running my own test, using the CCRL 40/4 testing methods, calculating with BayesELO, and I calibrate against CCRL:
Intel i7-6700K (crafty 19.17 bh: 17 seconds, or 40 moves in 85 seconds)
256 MB Hash per engine
Ponder off
Endgame Tablebase: 5 piece Syzygy
Book: Varied.bin, set to 8 moves deep
GUI: CuteChess, running 4 games at once (one per core, which is possible due to ponder being off.)
Having all levels play one another is useless. Level 20 will defeat even 19 about 99% percent of the time. Therefore I've decided to do it like this:
- Run a tournament between 20 to 16
- Run a tournament between 16 to 12
and so on.
Later, I'll sprinkle some other engines in between the larger ELO gaps and have them run a tournament against the two Stockfish levels above and below.
Below are partial results. ELO-ratings have been calculated from referenced PGN (CCRL forum: 720 kB is too large to attach here), and the list has been calibrated against Stockfish 10 x64, single-threaded, which is 3495 in CCRL 40/4. I will add the last few levels in the coming days, and then start to use some other engines to fill the gaps. I'll recalibrate the list against that "basket of engines" using the average ratings of those engines.
Code: Select all
Rank Name Elo + - games score oppo. draws
1 Stockfish 10 x64 1T20 3495 102 102 400 99% 2805 2%
2 Stockfish 10 x64 1T19 2855 32 32 400 45% 2965 26%
3 Stockfish 10 x64 1T18 2825 32 32 400 41% 2972 27%
4 Stockfish 10 x64 1T17 2793 32 32 400 36% 2980 27%
5 Stockfish 10 x64 1T16 2745 23 23 800 50% 2788 24%
6 Stockfish 10 x64 1T15 2677 30 30 400 60% 2602 26%
7 Stockfish 10 x64 1T14 2652 30 30 400 56% 2608 26%
8 Stockfish 10 x64 1T13 2552 31 31 400 39% 2633 21%
9 Stockfish 10 x64 1T12 2458 24 24 800 50% 2444 14%
10 Stockfish 10 x64 1T11 2371 33 33 400 65% 2253 14%
11 Stockfish 10 x64 1T10 2269 32 32 400 49% 2279 12%
12 Stockfish 10 x64 1T09 2181 33 33 400 36% 2301 13%
13 Stockfish 10 x64 1T08 2105 25 25 800 49% 2108 8%
14 Stockfish 10 x64 1T07 2050 34 34 400 66% 1910 7%
15 Stockfish 10 x64 1T06 1951 34 34 400 51% 1935 5%
16 Stockfish 10 x64 1T05 1863 34 34 400 39% 1957 4%
17 Stockfish 10 x64 1T04 1723 38 38 400 21% 1992 4%
If you wish, you could include "Stockfish 10 x64 Level 19" and lower in your list as "derivatives" of Stockfish 10 x64 in the single-core CPU list.
I hope this is of some use to someone. (edit: added the games up until now, levels 20 up to and including 08.)