[net.micro.68k] Integrated MMUs versus separate ones

gnu@sun.uucp (John Gilmore) (05/09/85)

> Gee, maybe some day Motorola will catch up to Intel and integrate the
> CPU and the MMU.
> 
> expecting lots of flames on this one...
>  Phil Ngai

Here's a little candle then, Phil...  It is certainly a pain for people
porting Unix systems, as well as those trying to come up with a 68000
Unix-binary standard, that there is no decent standard MMU for the
68000 series.  Motorola is fixing that, though I believe it is lower
priority than either the 68020 CPU or the 68881 floatbox, since most
customers already have an MMU but don't have a 32-bit processor or a
well-integrated float box.

Just think if they'd integrated the 68451 loser onto the 68010 -- we'd
all be stuck with it.  Instead, people were free to come up with their
own idea of how to build an MMU.  We at Sun have a no-wait-state
TTL&RAM MMU that we're very happy with, though it costs us in board
area.  Motorola is building one (68851 PMMU) that snarfs the page table
entries out of RAM into a TLB like a VAX, but it adds a wait state
unless you walk a verry narrow path (so I hear), and TLB misses are
expensive.  Signetics is building a great MMU (the 68920 MAC) that runs
no wait states, fetches page table entries from RAM, deals with
multiprocessors and local/global memories, and also provides complete
control of a cache built out of external static RAMs.  Its problem is
that it won't run past 12-16MHz and will not be out for a while.
Various other companies (Apple Lisa, Fortune, ...) have designed their
own idea of a good MMU.

The point is that the customer gets lots of options and can trade off
time-to-market vs cost vs features vs risk vs software work.

Out of curiosity, is anyone building National xx0xx based systems
using any MMU except the National MMU?  Or was it good enough that it
stuck as a standard?

doug@terak.UUCP (Doug Pardee) (05/11/85)

> Out of curiosity, is anyone building National xx0xx based systems
> using any MMU except the National MMU?  Or was it good enough that it
> stuck as a standard?

The NatSemi MMU interface is tailored so specifically to the 32082 MMU
chip that if you tried to build a custom MMU, you'd basically end up
with a functional duplicate of the '082.  The '082 is tolerable, so...
-- 
Doug Pardee -- Terak Corp. -- !{ihnp4,seismo,decvax}!noao!terak!doug
               ^^^^^--- soon to be CalComp