Re: Innovative Systems Floating Point Engine Rev B 1988

[Mike Westerfield <MikeW50@aol.com>]

On Aug 7, 3:15 am, "Michael J. Mahon" <mjma...@aol.com> wrote:
> Mike Westerfield wrote:
> > On Aug 4, 7:19 am, "Yvan" <yvan_d...@gmx.net> wrote:
> >> Hi,
>
> >> I found a Innovative Systems Floating Point Engine Rev B 1988.
>
> >> First, what is it? Can I use it in my IIgs? Do I need any software or
> >> drivers? Are there any documentations anywhere?
>
> >> Cheers
> >> Yvan
>
> > Hi Yvan,
>
> > There is a software patch that allows this card to intercept and
> > handle any calculation normally routed through the SANE numerics
> > package. That works pretty well, but there is significant overhead--
> > enough that some calculations are about as fast in software. (Not all,
> > and not the more complex calculations.)
>
> That's been my experience with math coprocessors.
>
> Trig functions and SQR are big wins (since their software
> implementations are so slow), but the simple ops, like add
> and subtract, are at best a wash.
>
> > There are also compiler directives for the ORCA languages that will
> > bypass SANE and tap directly into the card itself. This is
> > considerably more efficient, and works for long integer math as well
> > as floating-point. It turned out that short integer math takes less
> > time in software than the overhead of using the card, so the ORCA
> > languages do not use the card for short integers.
>
> > All in all, it's a great card if you intend to do any number-
> > crunching, especially in software you write. Congratulations on
> > finding one.
>
> The problem with most math coprocessors is that they require lots of
> overhead moving operands in and out of the card.  A DMA-based
> coprocessor might help a lot more.
>
> In comparisons done on a //e, I find that a math-heavy mix of operations
> are speeded up by a math coprocessor just about as much as by an 8MHz
> Zip Chip--with which most math coprocessors are incompatible (because
> they use the RDY line to stall the 6502 and the Zip Chip doesn't
> implement it).
>
> Since the Zip Chip speeds up everything and the math coprocessor only
> speeds up multiply, divide, and functions, the Zip Chip wins.  ;-)
>
> I don't know whether this math coprocessor is compatible with a IIgs
> accelerator, but that would be something to check...
>
> -michael
>
> AppleCrate: An Apple II "blade server"!
> Home page:  http://members.aol.com/MJMahon/
>
> "The wastebasket is our most important design
> tool--and it's seriously underused."

This coprocessor works fine with an accelerator card.

Whether it will speed up your calculations significantly is another
matter. It will certainly speed up simple floating-point calculations
like +, -, * and /, but the big win is if you are doing the more
complex calculations, like trig or other complex functions.

Mike Westerfield