Need buying advice for 386 and Unix

[James Van Artsdalen]

Quick preface: I don't argue that you can't buy SIMMs third party.  As
long as you stick with memory the manufacturer specifies you're OK
(assuming the SIMMs weren't sorted - an unusual thing for SIMMs I
hope).  But buying Joe's Discount 80ns DRAM may or may not work -
there may be good reason that the manufacturer doesn't use it.

In <5763 at crash.cts.com>, jca at pnet01.cts.com (John C. Archambeau) wrote:

> Why Toshiba?  Memory is memory.  NEC, TI, et. al.

| No.  [...]  Look at a DRAM specification sheet some day.  There are
| at least twenty different parameters that must be met.

> I have yet to find a memory module that you can't plug in.

These chips have nanosecond tolerances of many different types.  A
specification with a 30% tolerance may mean 13ns instead of 10ns.
Just because it "plugs in" and passes POST tests while cold doesn't
mean that you might not be eating deep into margin in a few minutes as
the box heats up.

> My 386SX motherboard manual lists a long list of memory that has
> been tested to work by manufacturer and their part number.

YES!!!  That's because *those* SIMMs have been tested and work.  If
the vendor knew that every SIMM would work, why bother printing this
list?

> But the chances of pulling a particular DRAM chip off the shelf and
> having it work are VERY high if it's a well designed board.

I claim the opposite result: a well designed board will accept fewer
types of SIMMs than a poorly designed board.  If you consider speed or
wait states a design goal, then you will design for tighter tolerances
to eliminate wait states.  Those tighter tolerances mean that fewer
SIMMs meet your specification.

A design that trades away performance for cost will have more wait
states.  This is not only easier & cheaper to design, but it means
that more SIMMs will meet specification, and hence the memory cost is
less.

Carried to an extreme, this is a bad idea of course: a design with
only one usable SIMM is no good from a manufacturing standpoint.

> Remember the memory chip shortage not to long ago?  A lot of chip
> vendors just plain weren't selling in the USA, so you took what you
> could get.

After it passed engineering design check, systems validation, and
environmental testing...

At that time, a lot of people went back and tried to loosen design
constraints.  Maybe the specifications on the memory controller had
tightened up since the design was shipped, maybe you could use a
faster PAL: lots of things could let you loosen the SIMM requirements.
-- 
James R. Van Artsdalen          james at bigtex.cactus.org   "Live Free or Die"
Dell Computer Co    9505 Arboretum Blvd Austin TX 78759         512-338-8789