Re: Apple ][ forever!

[dempson@actrix.gen.nz (David Empson)]

Jonathan Schattke <wizwom@ripco.com> wrote:

> These are the sorts of crazy ideas that float through my head in the
> middle of the night, and come out when my newborn son wakes me before
> they are properly consigned to oblivion.
> 
> I was thinking (oh no, not again!) that it was not only possible and
> plausible, but desirable to make a _fast_ Apple ][gs machine, with a
> fully extensible and compartmented architechture.
> 
> Modern RAM is 60ns - which allows 33 MHz operation.

Only with a processor which has an appropriate memory access cycle.  For
the 6502 family (including the '816), it would allow a maximum CPU speed
in the order of 8 MHz.  You would have to add external caching hardware
to achieve better than that.  Even just a multi-byte wide prefetch
scheme for read cycles would help considerably.

The 6502 family performs one memory access every clock cycle, unlike
other families like the x86 and 68K, which use multiple cycles per
memory access (at least for the older members - it gets really confusing
when you get to higher end models with caching).

The extra cycles allow increased time between presenting the address and
needing the data to be returned (or for completing a write operation).

> The precessor series used in the gs goes up to 14 MHz, right?

You would need 35 ns RAM to do a direct 65816-to-memory connection at 14
MHz.

I'm going on the basis of half the CPU cycle being required for memory
access.  It can actually use more like two thirds of the cycle, as the
address is available before the midpoint.  This reduces the speed
requirement somewhat: more like 45 ns, but it would eliminate the extra
time available for interleaving anything else like DMA.

-- 
David Empson
dempson@actrix.gen.nz
Snail mail: P.O. Box 27-103, Wellington, New Zealand