🚀 go-pugleaf

On 1 Aug 92, aki@cruzio.santa-cruz.ca.us wrote:
 
> I posted a question regarding the identification of a 386SX CPU
> versus a 386DX a couple of weeks ago.  I finally talked to one of
> the designers and found that the hardest way is the only way:
> 
> 1. Time a bunch of NOP's (don't loop'em)
> 2. Time a bunch of DWORD memory accesses (don't loop)

Hmm... that seems like the _really_ hard way. I always thought that
the prefetch queue length on the 386dx and 386sx also differed.  The
prefetch queue length can be measured relatively easy with a loop
which places an instruction in a bunch of NOPs just a little ahead
of it.

I've written a simple queue length calculation program a while ago,
I'd be happy to mail it to some people with 386sx or 386dx machines
(in fact, to anyone :-), and we could compare results then.
 
IMHO this is a slightly easier way to detect the sx. And wasn't IBM
coming out with a 386sx with 1K of cache these days? You can bet at
least the cache interface is 32bits, so you'd have to clean up the
cache first, before doing the above mentioned timing test (yes, they
might have changed the prefetch queue length too, of course).
 
-- Jan-Pieter
cornet@duteca.et.tudelft.nl