Image 01


Derek O'Connor Donard, Ireland

Science by ariya 36 comments

The link to Ng's paper should be - Oct 08 2006

Science by ariya 36 comments

I have just downloaded SpeedCrunch Version 0.7 Beta-1 and performed the following tests, using 50-digit precision:

SPCR sin(10^22 )= -8.52200 | 918050626935619363157432839453863959329520531e-1
PariGP sin(10^22) = -8.52200 | 84976718880177270589375302936826176215041004e-1
SPCR cos(10^22) = +5.23214 | 674176545467019501300134952278548556645769073e-1
PariGP cos(10^22) = +5.23214 | 78539513894549759447338470949214091997243939e-1

As you can see, SpeedCrunch performs very badly. It may have high precision, but it has only 6-digit accuracy in these tests. By the way, the standard Windows 2000 desk-top calculator has 32-digit accuracy.

If you want to see how difficult it is to write real numerical software, have a look at this paper : - Sep 04 2006