Author: Dann Corbit
Date: 19:18:04 09/16/05
Go up one level in this thread
On September 16, 2005 at 21:06:02, David H. McClain wrote: >On September 16, 2005 at 14:14:34, Dann Corbit wrote: > >>On September 16, 2005 at 12:52:13, Êèðèëë Îñåëåäåö wrote: >> >>>Hi, I am curious how much stronger a faster computer will be for the same >>>program. Is there any formula for a program rating change depending on the >>>system? >> >>A rough rule of thumb is 50-70 Elo for each doubling of CPU speed. >> >>But there is a lot of debate as to how well the rule holds, especially at >>extreme depths. >> >>So you would really have to carefully test to know the answer. > >Dann, > >An amateur's question: > >Is the CPU speed doubling theory related in any way to kn/s doubling estimates >for ELO gain? > >Example: > >Opponent: Toga II 1.0; 17.2 ply; 879 kn/s; Authentic AMD 2177 mhz (1 thread) > >Me: Shredder 7.04; 13.6 ply; 198 kn/s; 996 mhz; Authentic Intel PIII > >As you can see Toga is more than 4 times my kn/s but only 2.2 times my speed on >mhz. I understand that AMD XP architecture is perhaps not a proper comparison >to and older Intel PIII architecture. Is the theory still valid in this case? You cannot compare two programs. They do not even count the nodes the same way and if they did, the selection of which nodes to examine is more important than how fast you look at them. It is easy to make a super-fast program that only looks at wood count but it will play very bad. To see a speedup for hardware you must measure the same version of the same program or the comparison is almost meaningless.
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.