Author: Uri Blass
Date: 15:24:08 01/28/01
Go up one level in this thread
On January 28, 2001 at 17:56:06, Jorge Pichard wrote: >On January 28, 2001 at 16:58:22, Uri Blass wrote: > >>On January 28, 2001 at 14:33:23, Jorge Pichard wrote: >> >>>Six Weeks ago I matched Nimzo 8 Vs Junior 6 at G\60 using an AMD Athlon 800 MHz >>>And the score after 75 games ended up with a slight advantage of 6.5 points for >>>Nimzo 8. Then one week after that match, I decided to match the same two >>>programs, but I used an AMD K6-2 500 MHz instead, and the score after 75 games >>>ended this time in favor of Junior 6 by almost the same margin as the previous >>>match. I realized that certain program benefit more than others as the speed of >>>the processor increases, but I wasn't satified yet and decided to test the same >>>two programs one more time with a slower processor this time. I asked my friend >>>John to test these two programs again with his old Pentium Celeron 333 MHz at >>>G\60 and after 9 games, Junior 6 was beating Nimzo 8 by W7 D2 L0 and, at that >>>moment I decided to stop the match. I can only conclude that Nimzo 8 benefit >>>more as the speed of the processor increase, therefore, Nimzo 8 will not have a >>>great SSDF rating by the middle of February, but if you have a P.C. with a >>>processors higher than 800 MHz Nimzo 8 is one of the few programs that benefit >>>the most by using the latest technology available. >>> >>>Pichard. >> >>I do not understand why do you work so hard in playing games. >> >>I think that you should find the number of nodes per seconf of nimzo8 and >>Junior6 on: >> >>1) Celeron 333 >>2) K6-2 500 >>3)Athlon 800 >> >>For example >>If you find that nimzo's number are 250000,500000,1000000 >>when Junior's numbers are 300000,500000,830000 then it will be an excellent >>proof that Nimzo earns more from the new processors. > >The NPS is not the a true scientific measure of Knowing why a certain program >benefit more than another from the gain in processors speed. For instance, if >you provide Nimzo 8 to play a game by using 1 Minute per moves and at the same >time you provide Junior 6 also 1 minute per move. You might have this escenario: > >1. Let say that after one minute Nimzo 8 was able to reach XXXXXX NPS by using >the Celeron 333 MHz but was not able to find the best move within that horizon, >whereas Junior 6 by reaching XXXXXX NPS was able to find a better move. Now when >you increase the speed of the processor to a higher level (K6-2 500 MHz) Nimzo 8 >is now getting closer to zero in, where it could find a better move provided the >same amount of time of 1 minute per move, whereas Junior 6 is still providing >almost the same evaluation. Finally, when you increase the speed of the >processor to a minimum of 800 MHz, Junior 6 is of course calculating a higher >number of NPS, but is not improving the evaluation too significant as compared >to Nimzo 8 when is finally reaching is peak. The ssdf games are at 2 hours /40 moves that is more than twice slower than 1 hour/game I do not think that having p3-800 can compensate for the difference in nps. It means that if there is a problem with the ssdf results of nimzo then your results suggest that the problem is different than the case that nimzo earns more from slower time control on the same hardware. The only possible problem that I can think of except the possible case that Nimzo earns more nodes per second from faster hardware is the case when the meaning of nps in faster hardware is not the same and Nimzo can see with the same number of nps more if it has faster hardware when it is the opposite for Junior. It is hard to check it for Junior because Junior do different jumps at different time controls and it may jump from iteration 14 to 16 using hardware A when it jumps from depth 15 to 17 using hardware B. I do not know if it is possible to translate a time control for Junior with hardware A to another time control with hardware B in order to force Junior to do the same jumps. 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.