Author: Robert Hyatt
Date: 20:03:56 12/09/99
Go up one level in this thread
On December 09, 1999 at 18:18:42, blass uri wrote: >On December 09, 1999 at 17:41:00, Robert Hyatt wrote: > > ><snipped> >>We now have two issues. Assuming you use a 64 bit hash signature, what is the >>probability that two different chess positions produce the same 64 bit hash >>signature? That would cause serious problems, because we store search results >>and identify these results by the 64 bit signature, not the full position >>description. If two different positions have the same signature, we would use >>search results in the wrong places.. very bad. But this is _very_ rare. > >When the hardware gets faster it will not be very rare and the question is if >crafty can lose at long time control with good hardware because of it. >Can better hardware be counter productive? > >Is it possible that crafty with hardware 1,000,000 times faster is going to do >stupid mistakes and lose because of it. > >I know that crafty did good results against Rebel in chess2010 but crafty used >only 1 processor in this games. > >Is it possible that in this time control(and in a few years in tournament time >control) more processors are counter productive? > >Uri It is an issue. I ran lots of tests for 4 billion nodes each a couple of years ago. 4 billion nodes isn't reachable by today's hardware, excepting DB of course. Even at 1M nodes per second, which is just a bit over my quad xeon 400 speed, that takes 4,000 seconds to hit. On a big alpha, I can hit well over 16M nodes per second, which drops that further, but nowhere near what we could expect in 40/2hr time controls. 1 unnoticed error per billion nodes was the value I got when I ran the test. The next question is what is the probability that one error in 1 billion nodes will affect the root move/score? I personally don't think this will be a problem for many many years... But I wouldn't want to be depending on less that 64 bits. 32 was proven to be useless. 48 is now dangerous.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.