So the front page of yahoo says the newest fastest computer in China can perform 2,507 trillion calculations per second…Thats 2.5 quadrillion or 150 quardillion per minute. So let this computer think about the first move for 1 year and it’s going to think through 79 million quadrillion possible outcomes.
So I’ve lost touch but if (Rybka) was set-up on this machine how would the increased speed relate to the approx. 3000 rating she has on my regular single core home computer?
Also, how many moves deep could this “solve” after 1 year?
I can’t say how many moves in advance Rybka could think on the world’s fastest super computer, but its logrithmic. So each additional ply would take more and more calculation than the ply before it. So it would probably dissapoint you in how little it could actually calculate entire ply’s deep. (Obviously it would go very deep in more optomistic lines.) Maybe it can go 16 ply’s deep, with optimal lines going 5 to 10 moves deeper.
I think at best, Rybka might have say, 3600 to 3650 elo running on the super computer.
Not only is it logrithmic to go plys deeper, but you also get less performance per core as you add more cores.
Considering there’s never been a player with a 3000 rating and not that many with 2900 ratings, claims of a 3600 or higher rating for a chess computer have always struck me as unrealistic.
What’s the justification for adding 400 Elo points to Rybka’s current rating? At last report Rybka was tested on 52 cores (monster!). Processing faster doesn’t necessarily make the program smarter.
More likely the question is unanswerable because we don’t know how much space is available for the hash tables. Then you’d have to figure out whether you’ll feed it positions with forcing lines or make it attempt to solve something like Larsen’s opening , and how much processing time you’ll give it on any given position.
I figured 400 is about the max you could get with the number of cores. Mostly it comes down to the computer being able to draw (mostly), and win (sometimes) nearly every game, since no other computer would have nearly the processing power, it should be able to at least see the most probable line for every opponent engine.
I think its rate of losses would be very small indeed. But I think many games would still be drawn. Even a 6pct win rate, but nearly zero losses would eventually boost its long term rating quite a bit.
Yeah I know that if we doubled the current fastest computer that we might only see 1-2 plys deeper. 3600 did seem high. BUT I’m not sure of Rybka’s win % now but a better (i.e., faster) Rybka would lose less often bosting its rating above the other weakling 2900-3020 elo programs.
Still a cool topic…I remember when deep thought was a monster at 1 or 2 million positions per sec and all non-chess fans simply thought it would decimate Kasparov.
, and how much processing time you’ll give it on any given position.