1 Stockfish 1.3 JA 2 +308 =467 -224 54.20% 541.5/999 +29 ELO
2 Stockfish 1.2 (default) JA +224 =467 -308 45.80% 457.5/999 -29 ELO
So the regression, if any, or shows only at longer time controls or only against other engines....but it does not show for CPU 1 case where instead we are stronger then 1.2 also against other engines.
I am puzzled by these numbers and I would like to ask testing expert's an opinion on theese and on the corresponding CEGT results that instead shows a regression starting from CPU 2 case.
It doesnt make sense to run matches between 2 version of engines, one with single cpu and another 2 cpu. Better try Stockfish 1cpu vs Fruit and Stockfish 2cpu vs Fruit and compare the results difference.
swami wrote:It doesnt make sense to run matches between 2 version of engines, one with single cpu and another 2 cpu. Better try Stockfish 1cpu vs Fruit and Stockfish 2cpu vs Fruit and compare the results difference.
Perhaps I wasn't clear enough in my post. Title is wrong actually it should have been:
Stockfish 1.3 2 CPU vs Stockfish 1.2 2 CPU
But the macth was against 2 CPU version for both engines 1.2 and 1.3 run on an Intel dual core.
I agree that it has more sense to test against different engines but the point is that in CPU 1 case CEGT _seems_ to show an improvment against other engines too, while in CPU 2 case CEGT shows a regression not only against other engines _but_ also against 1.2 version, that's why I have directly tested against 1.2 in the CPU 2 case.
Just to be more clear, my not proven and possibly very wrong feeling is that Stockfish 1.2 is _understimated_ in CPU 1 case (where the difference from Glaurung 2.2 is too small) and more or less correctly or even over estimated in SMP case.... 1.3 perhaps is a little bit underestimated in SMP (because its ELO is less then 1.2 while actually it is very probably to be stronger also in SMP) and _hopefully_ correctly estimated in CPU 1 case.
This is of course just a very very weak idea and I trust CEGT results much more then my "ideas".
swami wrote:It doesnt make sense to run matches between 2 version of engines, one with single cpu and another 2 cpu. Better try Stockfish 1cpu vs Fruit and Stockfish 2cpu vs Fruit and compare the results difference.
Perhaps I wasn't clear enough in my post. Title is wrong actually it should have been:
Stockfish 1.3 2 CPU vs Stockfish 1.2 2 CPU
But the macth was against 2 CPU version for both engines 1.2 and 1.3 run on an Intel dual core.
I agree that it has more sense to test against different engines but the point is that in CPU 1 case CEGT _seems_ to show an improvment against other engines too, while in CPU 2 case CEGT shows a regression not only against other engines _but_ also against 1.2 version, that's why I have directly tested against 1.2 in the CPU 2 case.
I do not see where do you find CEGT shows regression against the 1.2 version because I see no match of 1.3 against 1.2 in the CEGT.
The only result that I can find of stockfish against previous version is
correctly estimated in CPU 1 case.