On Wed, 27 Aug 1997 20:58:12 +1000 (EST),
  Karl Maftoum <k.maftoum@student.canberra.edu.au> wrote:


>Hmm, this is important because the machines are being used for a 
>distributed computing experiment (using pvm) to solve mathematical 
>models, so the FPU sorta has to work properly! I'll investigate the 
>kernel code. BTW there is a BIG difference .. about 6 times on a float 
>multiply, how good is the maths library?



There is a simple test you could try:
    1. Compile and link a simple FP_intensive program under Ultrix.
	  (hopeful y a small-memory, FP-intensive  FP benchmark.)

    2. Time the program under Ultrix (preferably on an
       otherwise-unloaded system).

    3. Time the same program running on NetBSD 
       (i.e,. running the Ultrix binary usign Ultrix emulation).

    4. Compare results of 2 and 3....

That will determine if it's a library problem or a kernel
problem.