The /MT compile of Firenzina 2.2.2 xTreme that I released yesterday (03/01/2013) at http://dgusev.cs.edinboro.edu/GameDev/C ... nzina.html tied a 500-game 1/1 blitz match with the initial /MD compile released (02/25/2013), +75-75=350.
500 games played / Tournament is finished
Name of the tournament: FirenzinaMatch2MDvMT_KLO250_1-1
Site/ Country: DMITRI-ASUS, United States
Level: Blitz 1/1
Hardware: Intel(R) Core(TM) i7 CPU Q 720 @ 1.60GHz with 5.9 GB Memory
Operating system: Windows 7 Professional Professional Service Pack 1 (Build 7601) 64 bit
Some drawish result WAS expected, I am just shocked to see such round numbers! The test suite was KLO250. Now I am ready to try more suggested compiler options in addition to the /MT option. (/GS- was already in, and most runtime checks were off all along.)
I have completed a 200-game 3/3 blitz run of the newest optimized MT_SGVAD compile of Firenzina 2.2.2 xTreme, a clone of Fire 2.2 xTreme by Kranium (Norman Schmidt), against the released MT compile. The test suite was KLO100. SGVAD stands for Schmidt-Gusev-Velasco-Ablett-Dart. It's an abbreviation of last names of those involved in coming up with this set of parameters.
The MT_SGVAD beat MT, 103.5:96.5 (+27-20=153). According to the online ELO calculator, this represents a gain of slightly more than 30 ELO points. We may be near the ceiling of what can be achieved by playing with compiler settings. At this point, I intend to release this baby and proceed with making and testing generic x64 and x32 compiles for Windows.
In this case, the new compile cannot be worse than the old one, because (a) kN/sec are visibly higher, (b) the code didn't change, and (c) Four pros (Schmidt, Velasco, Ablett, and Dart) gave me advice on how to compile Firenzina better. I greatly appreciate that!
In this case, the new compile cannot be worse than the old one, because (a) kN/sec are visibly higher, (b) the code didn't change, and (c) Four pros (Schmidt, Velasco, Ablett, and Dart) gave me advice on how to compile Firenzina better. I greatly appreciate that!
This one gives +12, and so does this other one that I just found: http://www.3dkingdoms.com/chess/elo.htm
I was looking at the Elo rating change (starting at 3120 for both compiles and setting K=10), not for Elo difference. I used this calculator: http://www.kosteniuk.com/EloCalc/elo.php However, here also, if I plug in 3132 and 3120, then it expects 103.4 points out of 200 for the stronger player.
Gusev wrote:I have completed a 200-game 3/3 blitz run of the newest optimized MT_SGVAD compile of Firenzina 2.2.2 xTreme, a clone of Fire 2.2 xTreme by Kranium (Norman Schmidt), against the released MT compile. The test suite was KLO100. SGVAD stands for Schmidt-Gusev-Velasco-Ablett-Dart. It's an abbreviation of last names of those involved in coming up with this set of parameters.
The MT_SGVAD beat MT, 103.5:96.5 (+27-20=153). According to the online ELO calculator, this represents a gain of slightly more than 30 ELO points. We may be near the ceiling of what can be achieved by playing with compiler settings. At this point, I intend to release this baby and proceed with making and testing generic x64 and x32 compiles for Windows.
Dmitri, if it were me I would do it the easy way. You have 2 %s: .5175 and .4825. Put your system's calculator on "scientific" and then do this:
.5175 / .4825 = log * 400 = 12, with 12 being the elo difference.
There are a number of ways- this is just the simplest for me.
Gusev wrote:
In this case, the new compile cannot be worse than the old one, because (a) kN/sec are visibly higher
If the new compile's search tree is identical to the original Fire 2.2 compile (i.e. node count, pv, selective depth, root move ordering etc. - all of which may be seen from UCI output - are always identical), then the elo gain is most easily and accurately measured by taking the percentage difference in speed (kN/s).
If you run a testset of positions, it should be possible to get the approximate speed increase and hence the elo gain.
