Re: Number Cruncher FPUs

[dinarich@di.epfl.ch (Mirko Dinarich)]

In article <4olr9t$l6a@trotsky.cig.mot.com>, chinay@cig.mot.com (Albert Chin-A-Young) writes:
> Also, the 25Mhz won't do you much good. Due to the limited speed
>of the bus for the II, the 25Mhz won't buy you much. I have a 16Mhz in
>mine and that's probably overkill. The 12.5 or 16.67 should be
>sufficient.
>-- 
>albert chin (ITS Service Team 2)

Well, it depends because this kind of fpu takes a great number of cpu cycles
in order to achieve most floating instructions .
Here is some instruction's execution time in clock cycles (taken from MC68881 Floating-Point Coprocessor User's Manual):

FABS  650    (PS: these # are without effective address calculation time and
FASIN 606         are suitable for double calculation)
FCOS  416    
FDIV  130
FLOGN 550
FMUL   98

Even the fastest 68881/2 will not finish most instructions within 1 Mhz master
clock cycle. 
I don't know how works the Number Cruncher but IMHO the fastest fpu get the
quickest results even with the 1 Mhz bus bottleneck.

-- 
!\  /!  !  !  /  /--\  /--\  /--\  /--\     
! \/ !  !  ! /   !  /  !  !  !  /  !  !  dinarich@di.epfl.ch 
!    !  !  ! \   ! \   !  !  !  \  !  !  http://diwww.epfl.ch/~dinarich
!    !  !  !  \  !  \  \--/  \--/  \--/