Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka on Blue Gene
- - By ChessMate (***) Date 2007-07-27 20:48
Its interesting to know  that If we managed to run Rybka 2.3.2a  on   " Blue Gene  " ( a 360 teraflops machine )  then what would be :

1. Its relative human elo (assuming atleat 1000 games against 2600 - 2800+ human players at standard time controls ?

2. How much ply it can get in standard average 1 move time ?

3. What would be the results if there are 100 games at tournament time controls  between   Top human  (Anand, Topalov , Vladimir Kraminik or Garry Kasparov)  assisted with rybka 2.3.2a  on good 8 core machine  vs   Rybka 2.3.2a running on Blue Gene with a strong opeining book  ?
Parent - - By boo! (**) Date 2007-07-27 21:27
Blue Gene consists of something like 100 000 processors. The "chess problem" is not very well suited for such a high number of parallel processing units. Basically you reach a point where adding more processors adds nothing to, or actually reduces, the nps count. For rybka this point lies well below 50 processors!

With a dts-implementation(a different way of paralellising the chess tree), this number would be significantly higher, but probably no where near 100 000.
Parent - By Vasik Rajlich (Silver) Date 2007-07-28 08:43
It's hard to speculate about massive hardware like this. On paper, a Blue Gene Rybka would destroy any centaur on normal 8-core hardware.

Re. DTS - actually, I've been thinking about this the past week. DTS is not really the way to go. Parallel search in computer chess is presently a terribly backward area.

Vas
Parent - By Gr00vy (**) Date 2007-07-28 01:47
It really isn't the case that this will dramatically improve ELO at the rate you think.

The deal with Rybka, is that it makes better moves at Ply 1.  That it's initial choices are simply better, and it already extends search very well.  So as the depth gets deeper, it is going to be less and less likely that you are going to find a horizon quality type tactics that are otherwise invisible that are going to score you so poorly that you have to change the move.

Rybka changes the game, because precisely it plays better chess.  This is a bit different than the previous race, which was to play the least bad chess as deep as you can.   Which actually worked really well, up to a certain depth.  After a certain depth, it is too difficult to get to that next ply because of the exponential nature of that problem.  

There are still many things that Vas would like to work on before we get to Depth as a problem.  And as I understand it, narrowness and better play, not least bad play, is the hallmark of this engine.

And ultimately, since solving chess is not on the table, and beating the engine is no longer on the table, what is more interesting to me, is the quality of the Super-GM that is over my shoulder helping me understand the game, and Rybka is firmly in my arsenal!
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka on Blue Gene

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill