286 -> 386sx Upgrades
Dick Dunn
rcd at ico.isc.com
Wed Sep 26 08:44:38 AEST 1990
tuv at pmafire.UUCP (Mark Tovey) writes about upgrading a 286 AT to 386:
[> > is also Tovey]
> > I have a friend who has found a 386sx upgrade kit. Apparently you
> >remove the 286 chip and install a plug in daughter board in place.
> >This board contains the 386 chip, a socket for a coprocessor (you can
> >continue to use the 287 if desired) and any necessary support
> >circuitry.
I tried a couple different flavors of these and returned both. I've looked
into others and rejected them all for one reason or another. One thing
you've got to watch out for is whether the add-in board will fit the space
available. One board would almost fit, except for obstruction by a power-
supply connector which came up too high. I tried to raise the daughter-
board with a chip carrier; this introduced a noise problem and it wouldn't
work. Look at the size and layout of the board, then look at where your
286 sits (paying attention to location of pin 1). You'll need to get the
info for the right kind of CPU package (PLCC or PGA). Looking at the SOTA
ad, I noticed that they seem to have arranged their PGA board so it won't
clear a ZIF socket right! (A ZIF has a lever along one side to apply the
pressure to the contacts. This requires that a daughterboard mount the
pins to the old processor socket along an edge of the board.)
Most clones use PLCC instead of PGA; make sure the vendor of the board
offers the right connection. Also, if you're looking at this for a UNIX
machine, be *sure* you can return it if it doesn't work for you--and try it
out with UNIX right away. Often, add-in boards have only been tried with
DOS, and a support call will only get you a "What's a UNIX?" Also, there's
the problem of boards which aren't really hardware-compatible; they're
"BIOS-level" compatible--no use for UNIX, which doesn't use the BIOS.
> I managed to track my friend's source. He got the information from
> the August 20 issue of InfoWorld, page 83. The company that
> manufactures it is:
> SOTA Technology Inc.
[details deleted]
> It uses a 16MHz chip and comes with 16k of cache memory...
>...The cost is a little steep, it lists for $595.00. This may or may not
> be cheaper than replacing the motherboard and memory, depending on
> your configuration...
$595 is definitely steep. You should be able to get a complete SX mother-
board for $400 or less; that will leave you $200--with which you can go
out and buy 4 Mb of memory. (Well, 1Mx9 80ns SIMMs are almost down to
$50; they probably will be by the time this message gets out.:-)
Thoughts on relative advantages: The daughterboard has a cache. This will
help performance in some areas, *assuming* the cache works right (e.g.,
doesn't tangle with I/O ports, memory-mapped I/O devices, or DMA). It
will let you use existing (bus) memory, which would be good if you've
already got a lot. BUT if you've gotten a lot of bus memory, that might
have come about because you needed it (:-) for large programs and/or lots
of data--in which case the cache is likely to be inadequate. If you're
upgrading an old machine, consider that bus speed is probably 8 MHz tops;
that's a 125 ns cycle. An access to bus memory takes 3 cycles, or 375 ns.
per 16 bits. By comparison, if you replaced the entire motherboard, you'd
put normal 80-ns memory on it. Some trickery actually allows the 16-MHz
SX motherboards (that's 62.5 ns cycle) to use 80 ns memory without a wait
state on every access. Anyway, ignore the details; you should be getting
the idea: old slow bus memory plus a cache may lose against faster on-
board memory. And if you don't already have a lot of bus memory, memory
expansion is cheaper with the new motherboard.
--
Dick Dunn rcd at ico.isc.com -or- ico!rcd Boulder, CO (303)449-2870
...Worst-case analysis must never begin with "No one would ever want..."
More information about the Comp.unix.sysv386
mailing list