
Michael Hart wrote:
I keep asking for a better model of prediction,
If you want a 'doubling every 18 months' curve, then using a reference point of 100 in 1993 actually provides a much better fit to the (post-1990) data than does 10 in 1990.
... the failed suggested models of starting 1971 and 1993 as the best baselines for Moore's Law, ...
Re the "failure" of the 1993 model, in yesterday's posting: http://lists.pglaf.org/private.cgi/gutvol-d/2005-January/001419.html you gave this data: (omitting the columns that use 1971 and 1979 as reference points) START TOTAL START TOTAL YEAR NUMBER YEAR NUMBER ACTUAL 1990 BASE 1990 1993 BASE 1993 NUMBER/YR 1 1971 4 1974 7 1977 9 1980 9 1983 10 1986 1990 10 10 1989 1993 40 1993 100 42 1992 1996 160 1996 400 365 1995 1999 640 1999 1600 1550 1998 2002 2560 2002 25600 4260 2001 2005 10240 2005 102400 14944 2004 2008 40960 2008 409600 ????? 2007 However, there's a mistake in the "BASE 1993" column: the numbers after 1600 should be 6400, 25600, and 102400. Moreover, it's easier to make the comparison if we use the same years for the actual numbers as we do for the predicted numbers. With these two changes, I think the table should be something like: ref point = ref point = year '10 in 1990' '100 in 1993' actual 1990 10 25 10 1993 40 100 100 1996 160 400 750 1999 640 1600 2000 2002 2560 6400 6500 2005 10240 25600 20000? 2008 40960 102400 ????? (It should also be clarified that the number for a given year is that of Dec 10th for that year.) As you can see, the actual numbers (after 1990) are closer to the "1993" curve than the "1990" curve. -Michael Dyck ps: In your table, there's also a typo in the "BASE 1971" column: 1480576 should be 1048576.