Author: Uri Blass
Date: 15:35:00 01/28/01
Go up one level in this thread
On January 28, 2001 at 18:03:10, Severi Salminen wrote: >>I do not think that longer time control is the same as faster computer. > >In general case it is - this means that we forget how compiler optimizations >affect and other stuff. Because the processor speed determines how much of work >an engine can make in period of time, time also determines how much work can be >done. It is compitely different case if we take into account different processor >architectures and optimizations to different platforms. In these cases we can >not talk about speed but about different processors. > >>I will give an extreme example: >> >>If you give Crafty to play against nimzo8 when both programs use one processor >>system then it is going to do worse results relative to the results of it if you >>give both programs 4 processors. >>The reason is very simple >>Nimzo knows to use only one processor out of 4. > >Exactly. This is not about speed but about architecture. I was talking about architecture because architecture. There was no logical reason to get a conclusion about which program earns more from time. The only experiment to check which program earns more from time is to give the program the same hardware and different time control and not different hardwares. > >>Both programs earn from faster hardware and the case is not extreme as the >>previous case when one program earns nothing but it is possible that one of >them earns more. > >That is true, but I think we should separete two things: benefit from speed and >benefit from different architecture. And then there is the benefit in strenght >which can not be seen from increase in NPS. This is very important to realize. > >>The question is not which program is better but which program earns more from >>fast processors. >> >>I do not use the number of nodes per second to decide which program is better >>but the increase in % in the number of nodes per second to decide which program >>earn more from the fast processors. >> > >As I said it _only_ indicates how much more nodes a program can crunch with >better hardware. And yes, this might be different for different programs as they >have been compiled with different compilers and use different assembly language. >But it does not tell if the differnce between strenght grows in same way as >difference NPS. So if we see that when moving from Athlon 600 to Athlon 1.2 >Nimzo earns 50% more NPS and Crafty 100% (which sounds _very_ unlikely) we still >cannot say that Crafty's strenght increases by same amount compared to Nimzo. > >Severi We can probably say that Crafty will do better against nimzo in 1 hour per game on athlon1.2 and not in 2 hours per game on athlon600 so we can say that Crafty likes more the faster hardware when we do the right comparison. Uri
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.