| |
BGonline.org Forums
Bot Comparison after 60/500 games.
Posted By: neilkaz In Response To: Bot Comparison after 60/500 games. (Phil Simborg)
Date: Thursday, 5 November 2009, at 2:04 a.m.
The purpose of XGR+ as I proposed to Xavier and he loved the idea is to give users with high end computers options for a slower, yet stronger XGR.
My desired design goal was 5 ELO stronger and 4-5 times slower. Currently the new beta version is about 5.5 ELO stronger (Xavier can rather accurately determine that with his developmental software) and 4 times slower. .. ie XGR+ is about as strong as 5-ply and considerably faster.
Out of maybe 100 test positions, mostly cube decisions I've rolled out or seen posted online, XGR+ is outperforming XGR and there's only one where XGR+ drifts towards the wrong answer whereas XGR and a stat sig rollout say otherwise.
For those with i7 type comps (like 8 core) or even a fast dual core, (this LT is a quick but not blazing dual core) I recommend analysing matches using 4 ply and then XGR+ for the second pass (if overnight rollouts can do some 2nd pass stuff).
I think I could play a match vs XGR+ on my fast comp at work...2.4 Ghz server that is effectively 8 core at ABT time controls and maybe it wouldn't time out.
If it did time out I think it would survive 2min+ 15 sec delay, and that a brand new i7 the is about 3 Ghz and equiv. 8 cores could play ABT time controls. That makes for one superb opponent and at a very acceptable speed. One can see what I will be using for my training soon.
I can't speak for the developer, but my understanding is that the new update with the ability to support more than 2 cores and XGR+ is due in a couple of weeks.
.. neilkaz ..
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.