Path: utzoo!attcan!uunet!cs.utexas.edu!uwm.edu!rutgers!rochester!kodak!sisd!jeh From: jeh@sisd.kodak.com (Ed Hanway) Newsgroups: comp.sys.amiga Subject: Re: Amiga Competitiveness. Message-ID: <1990Oct1.132509.8967@sisd.kodak.com> Date: 1 Oct 90 13:25:09 GMT References: <1990Sep28.193008.6137@sisd.kodak.com> <1990Sep29.081216.17970@assari.tut.fi> Sender: news@sisd.kodak.com Organization: Printer Products Division Eastman Kodak Lines: 25 h112706@assari.tut.fi (Herranen Henrik) writes: > >In article <1990Sep28.193008.6137@sisd.kodak.com> jeh@sisd.kodak.com (Ed Hanway) writes: >>Yes, for an 8-hour ray tracing, saving one hour would be significant. >>Ray tracing, however, depends almost entirely on floating point speed. >>A 14MHz 68000 won't do a whole lot for that, but an add-on 68881 might. > >Doesn't floating point calculating use heavily the slowest commands in the >68000/68010, divide and multiply? I've noticed that after I inserted the >68020 in my B2000, many Mandelbrot programs speeded up with 20-25% while >AmigaBASIC only gained 5%. I'll concede that software floating-point emulation code does probably make heavy use of integer multiply and/or divide instructions, which should benefit from a faster CPU, even without correspondingly fast memory. These instructions also benefit from improved microcode in the 68010 and later chips. I stand by my original assertion that slight speedups are only noticeable for lengthy operations. It turns out that ray tracing and Mandelbrot calculations happen to be good examples of applications that do benefit from slight (<50%) speedups. Ed Hanway uunet!sisd!jeh