Author: Uri Blass
Date: 20:34:47 03/30/03
Go up one level in this thread
On March 30, 2003 at 22:05:47, Christophe Theron wrote: >On March 30, 2003 at 21:39:02, Mike Byrne wrote: > >>On March 30, 2003 at 13:35:07, Chris Carson wrote: >> >>>On March 30, 2003 at 11:53:21, James Constance wrote: >>> >>>>1. Why is there such a difference in Nodes Per Second in Pocket PCs when >>>>compared with PCs? e.g. Shredder on my Athlon 1200 Mhz might do 180,000 NPS for >>>>a given position, whereas my StrongArm 200 Mhz does only 4,500 NPS for the same >>>>position. That's a factor of 40 - I was optimistically expecting it to be >>>>closer to the ratio in processor speeds. >>>> >>>>2. Anybody have a stab at how much this reduces the SSDF quoted rating by? >>>> >>>>cheers! >>> >>>I collected NPS for a few Pocket PC's and Palm devices and also for the SW on >>>PC's. The 206 seems to be about the same NPS as a 33Mhz 486 and about the same >>>playing strength. For a specific program (ie Ruffian or Chess Tiger), the ELO >>>will drop by about 70 points for each 50% reduction in NPS. So a Pocket PC >>>206Mhz is about 400 points weaker than an AMD 1.2Ghz. Same for the Palm. Take >>>a look at the SSDF list: >>> >>>http://w1.859.telia.com/~u85924109/ssdf/list.htm >>> >>>Chess Tiger 15 AMD 1.2Ghz is 2726 SSDF >>>Chess Tiger 14.9 Palm m515 is 2101 SSDF >>> >>>Ruffian 1.01 AMD 1.2Ghx is 2671 SSDF >>>Ruffian 1.04 ARM 206Mhz is 2050 aprox. >>> >>>I have played all four of the above and this is also my experience. I am rated >>>2150. I have also posted NPS for the four above in previous posts as well as >>>Chess Genius (similiar experience on both machines for CG). >> >> >>Thanks for the post -- I disagree with the 70point reduction for 50% less power. >> The inverse of that , 70 point increase for 100% increase is clearly too high. >>I would cut the 70 point reduction in half. Say 35 point delta for every 100% >>increase or 50% reduction. > > > >Do it with actual data, and you will see that 70 elo points for doubling the >processor speed is exactly what the SSDF list gives. > >I have computed this 2 years ago from all the SSDF data I had. > >Maybe the numbers have changed now that we have faster processors (I believe in >dimishing returns eventually). > >But at least for the kind of processing power we have in handhelds now, the 70 >elo for a doubling of speed works extremely well. > >I have even published at my web site the mathematical formulas that give the elo >of Chess Tiger depending of the processor speed. > >You can find them at http://www.chesstiger.com (click on the FAQ link and look >for the "What is the ELO rating of Chess Tiger, and what does it means?" >question). > > > > Christophe Only one point about rating. The ssdf rating is not the same as fide rating. The list was increased by 100 elo. The original list was supposed to be equivalent to fide rating and they decided to reduce 100 elo for all programs to do the rating of the top programs more correct. I think that 2100 ssdf rating should be translated to 2200 fide rating unless the level of chess programs is improved. I think thatit is a pity that there are no weak chess program with fixed rating so humans always can get rating relative to the programs. I think that Fide should give some non deterministic machines to play in fide tournament when the machines are not going to be the best. one machine may have fixed rating of 2100 another machine may have fixed rating of 2500. All humans should be rated relative to these machines. If results suggest that the gap between the machines is not exactly 400 elo then it is easy to do some linear change of the rating of all humans to do it 400 elo. I think that this idea may help to get reliable rating with no inflation for humans. The machines should be private and not commercial so humans cannot buy them to try to find weaknesses that may distort the rating(learning about weaknesses from training is easier than learning about weaknesses from games and I do not expect humans to do special preperations against the non deterministic machine if they know that the machine is only one of a lot of opponents that they play. 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.