mcostalba wrote:we don't foreseen an important gain from 1.7.1, just a small one...
actually iirc the initial word was no gain in 1.7 but big gains after
btw, idk if you saw on the other forum but my match is finished, +22 elo for 1.8 not so far off from where you think the gain is, but i shd just forget hyperbullet eh?
Hyperbullet gives a (too) big premium to the fastest version and 1.8 is a bit faster then 1.7.1, but with longer TC this premium greately decreases.
mcostalba wrote:
There are no big new features, mainly code shuffle and cleanups, of course we expect an ELO increase, but this time we are quite confident
it will be smaller then last time
Early days in my testing, but I'm very impressed by what I've seen so far. Well done to the Stockfish team. Just love watching this engine play.
Taner Altinsoy wrote:
I guess the speed gain is like % 5-7. Am I right?
Correct speed difference can be measured only when there is no functional change bewteen 2 versions of the same engine. Otherwise the searched tree will be different and speed comparisons are no more reliable.
So it is not possible to correctly measure the speed difference bewteen 1.8 vs 1.7.1
Said this, I can also add that knowing the speed increment of each single patch that increased speed without changing functionality I would guess we are about at 4-5% of pure speed increase.
This seems a small value, and acutally is, but I can assure you that to squeeze some extra speed out of a mature engine like 1.7.1 while preserving or even incrementing code readibility it is no easy task
After Germany beat Argentina and heavy rains in our countries Ingo's room is now cooled down enough for continuation of Stockfish gauntlet:
Stockfish 1.8
Stockfish 1.8 JA - Deep Rybka 4 (2945) 36.0 - 49.0 42.35% Perf=2892
Stockfish 1.8 JA - Houdini 1.02 (2933) 35.5 - 48.5 42.26% Perf=2879
Stockfish 1.8 JA - Naum 4.2 (2815) 51.0 - 34.0 60.00% Perf=2885
Stockfish 1.8 JA - Komodo 1.2 JA (2803) 52.5 - 32.5 61.76% Perf=2886
Stockfish 1.8 JA - Deep Shredder 12 UCI 32b (2800) 56.0 - 27.0 67.47% Perf=2926
Stockfish 1.8 JA - Critter 0.70 (2786) 60.5 - 23.5 72.02% Perf=2950
Stockfish 1.8 JA - HIARCS 13.1 MP 32b (2734) 59.5 - 24.5 70.83% Perf=2888
Stockfish 1.8 JA - spark-0.4 (2711) 68.5 - 15.5 81.55% Perf=2969
Stockfish 1.8 JA - Zappa Mexico II (2710) 62.0 - 22.0 73.81% Perf=2889
Stockfish 1.8 JA - Protector 1.3.2 JA (2701) 69.0 - 14.0 83.13% Perf=2978
550.5 - 290.5 65.46% Perf= 2905
841 out of 1000 games played
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
^^ 1.8 is probably only about +20 ELO, but nevertheless it was a good idea to publish a new version. Bugs were fixed, redundant code was removed (for example queen contact mate detection in eval is gone!), many things were reorganized, the executables are noticable faster now. From what I've seen so far, I'm very happy with the 1.8 release. Good work! Germany vs. Argentina was ok, but most of the time I prefered to look at the 1.8 sources.
Ralph Stoesser wrote:1.8 is probably only about +20 ELO, but nevertheless it was a good idea to publish a new version. Bugs were fixed, redundant code was removed (for example queen contact mate detection in eval is gone!), many things were reorganized, the executables are noticable faster now. From what I've seen so far, I'm very happy with the 1.8 release. Good work! Germany vs. Argentina was ok, but most of the time I prefered to look at the 1.8 sources.
A 20 Elo improvement at the top is quite substantial, you guys have some unrealistic expectations. Congrats to the Stockfish team for continuing to improve their great engine!
And Germany was awesome this afternoon, made Argentina look like a very poor team (like England) .