PP likes to experimentlucasart wrote:you are comparing different compiles of the *same source code*, so you don't need to play any games. If there was a 1 elo difference between the two, you would need tens of thousands of games to detect it anyway.ThatsIt wrote:... after 600 games:
http://computerschach.forumfrei.net/t23 ... itions#392
Best wishes,
G.S.
just run the program from the command line, and type this:and compare:Code: Select all
uci ucinewgame position startpos go depth 20 quit
* node count, score and PV should be equal otherwise one compile at least is broken
* if so, then look at the time (displayed in ms)
and now you know *without any doubt* which compile is better, and you've saved your self a huge electricity bill of testing day and night for weeks...
sometimes his compiles are not changed in any substantial way from the ippolit.wikispaces.com released tarball
but he sometimes changes piece values, adds CPU detection routines, assembly language codepath (SSE, etc.) libraries., etc. and more
for ex:
http://www.talkchess.com/forum/viewtopi ... ht=#436990
there are in fact substantial changes, i.e. quite a few things that may/should affect ELO...and the way the engine plays.
KLO normally does not experiment, and as a whole has produced very consistent and accurate compiles with the vast majority of code intact.
he does add CPU detection, etc.
and recently has been changing/tweaking the TC routines...
the only way to know for sure if the codebase is 100% original is to ask, as neither compiler releases their 'changed' source code
Windows users are limited to these 2 choices: KLO's compiles or PeterPans
under these circumstances it makes complete sense to test them...i simply recommend asking 1st, especially if you are a 'purist' and want a Windows binary exactly as
intended by the authors.
PS-
Frank Q. ran 1400 games testing IvanHoe 46 (PP vs KLO)
and KLO ended up on top by 6 ELO.
