Author: Tina Long
Date: 14:33:50 01/08/02
Go up one level in this thread
On January 07, 2002 at 18:02:34, Uri Blass wrote: >On January 07, 2002 at 15:02:18, Robert Hyatt wrote: > >>On January 07, 2002 at 13:40:45, James T. Walker wrote: >> >>>On January 07, 2002 at 10:22:46, Robert Hyatt wrote: >>> >>>>On January 07, 2002 at 10:02:37, James T. Walker wrote: >>>> >>>>>It's more or less accepted that a doubling of speed gives only about 50-70 elo >>>>>points increase. Therefore your chart makes not sense. If we use a compromise >>>>>of say 60 elo increase the 2000Mhz machine would score approximately 58.5%. >>>>>Also there is no reason for this percentage to vary with time controls using the >>>>>same engine. >>>>>Jim >>>> >>>> >>>>Actually there is a _big_ reason why the data came out as it did. Look at >>>>anybody's results where they used the _same_ program, but played one copy at >>>>depth=N and another at depth N-1. At shallow depths, N wipes N-1 out. As the >>>>depth goes deeper, N does't do nearly as well. IE 4ply to 3ply, for the >>>>_same_ program, is a 33% deeper search for 4 ply. For 10 ply vs 11 ply, >>>>the difference is 10% deeper... >>> >>> >>>Hello Bob, >>>What is the point of your post? I don't see what it has to do with the thread. >>>Can you explain further? I like learning about stuff like this but you have >>>confused me. >>>Jim >> >> >>Adding one ply of search in a match X vs X will improve the results for >>that side. But if both programs can search to depth=3 and you add 1 ply >>to one of them, that is a _huge_ advantage. But if both can search to >>depth=14, then adding 1 ply to one is _not_ such a huge advantage. I simply >>pointed out that the results that were posted (2x faster hardware produced >>much more lopsided results as the games got faster and faster) were quite >>normal and expected... >> >>Giving one program 2x faster hardware is close to giving it one extra ply of >>search... > >The numbers that were posted do not make sense > >Here are the numbers that were posted: > >1sec 3sec 6sec 10sec 15 sec 20sec 30 sec 45sec 1min 2min 3min 5min > >95% 93% 91% 89% 87% 85% 83% 81% 79% 77% 75% 73% > > >The program that is twice faster is not going to get 95% even at 1 second per >move. > >I am not going to argue about the question if there is deminshing returns here >but if there is a diminishing return from depth then something like 62% at 1 >second per move and 56% at 5 min per move is more logical. > >Uri My initial thought was 60% across the board. After reading Dr Bob's post I changed my estimate to 65% descending to 55%. Tina Long
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.