Author: Roy Eassa
Date: 10:27:21 05/27/02
Go up one level in this thread
On May 27, 2002 at 13:11:42, Christophe Theron wrote: >On May 27, 2002 at 03:43:10, Ulrich Tuerke wrote: > >>On May 27, 2002 at 01:26:53, Uri Blass wrote: >> >>>On May 27, 2002 at 00:30:24, Frank Schneider wrote: >>> >>>>On May 26, 2002 at 18:49:23, Jorge Pichard wrote: >>>> >>>>>So far CT 14.9 is showing a SSDF rating less than 2125. >>>> >>>>IMHO computer vs. computer matches exaggerate the difference, I think >>>>Comet and Tiger were closer if both were matched against humans. >>>> >>>>However, the match shows that even a very good engine can't easily >>>>compensate a factor 10-20 hardware disadvantage. >>>> >>>>Frank >>> >>>The hardware advantage of Comet is a lot more than being 10-20 times faster. >> >>That's probably correct. >>I have heard that the difference from a Palm to a 486/50 is about a factot of 6. >>Accounting for another factor of 8 or so, you 'll get the P200. >>So, I'd guess Comet's advantage is roughly a factor of 50 in this match. >>Quite a a lot. >> >>Correct, Chris ? > > >Let's compute it another way. > >The Pentium Pro executes roughly one instruction per cycle (actually I think >it's 1.1 instruction/cycle in average). That makes 200M instruction/second. > >The DragonBall needs IIRC 14 clock cycles per instruction (average). Maybe >somebody can confirm or infirm this (the DragonBall is a 68000). > >So 48M/14=3.429M instruction/second. > >So the PP200 @200MHz seems to be approximately 200/3.429 = 58 times faster than >the DragonBall @48MHz. > >But it does not look right because on my Palm (m505 @54MHz) I get a TigerMark of >1.3 and on a K6-2 450MHz I get 225. So the TigerMark on PP200 should be close to >100 (which is actually how the TigerMark has been calibrated: PP200=100), and so >by this method PP200 is 77 times faster than DragonBall @54MHz, so PP200 is 87 >times faster than DragonBall @48MHz. > >Let's say it's between 58 and 87 times faster. My best guess: Comet's hardware >is 72 times faster than Tiger's. > I hate to throw in another factor, but having programmed in 8088 and 68000 assembly language, I think that many higher-level operations require fewer instructions on the 68000 than do equivalent operations on the 8088. This might mitigate the factor a bit (i.e., instead of 58-87, maybe it's more like 2/3 of that, or 39-58, which would then include the factor of 50 previously estimated).
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.