Replacement '386 motherboards ?

[Image Processig Center]

In article <24 at stanton.TCC.COM>, donegan at stanton.TCC.COM (Steven P. Donegan) writes:
> In article <354 at yunccn>, landolt at yunccn (J. Paul Landolt) writes:
> > In article <539 at bronson.UUCP>, tan at bronson.UUCP (N.R. "Tan" Bronson) writes:
> > > 
> > >     I would like to upgrade my PC limited 8mhz '286, running SCO 2.1.3,
> > > to a '386 running SC0 2.2.
> > >     The most attractive would appear to be to replace my motherboard.
> > 
> > [ The new motherboard, among other things, must be ]
> > >     - verified to work w/SCO xenix
> > 
> > Dyna SX386 - Dyna Computer Inc. $1,650
> >   
> > Hauppauge 386 Motherboard - Hauppauge Computer Works Inc. $1,495
> > 
> > Intel iSBC 386AT - Intel Corp.  sbout $1,995
> > 
> > Monolithic Microframe 386 - Monolithic Systems Corp. $3,295
> >   
> > Zeos 386 Motherboard - Zeos International Ltd. $1,395
> > 
> 
> Unless you have to have a 386 box for 386 Xenix I would highly suggest that
> you spend your money on a 286 board like this and a 287 to go with it.
> The company I bought mine from is: Computrader, 780 Montague Expressway
>                                    Suite 501, San Jose, Ca 95131
>                                    408-435-2662
> 
> -- 
I disagree with the above reccomendation. I've written an article which will
appear in Micro/Systems Journal this August in which I investigated the
performance of the three 386 UNIX ports (SCO XENIX, Microport, and Interactive
using the INTEL INBOARD 386 and a plain-jane 10mhz knockoff of the original
IBM 6 mhz AT motherboard. First, bear in mind that the 386 runs native mode
code twice as fast as emulated 286 code. Therefore, a 16mhz 286 running
SCO XENIX 286 will not be as fast as 16 mhz running XENIX 386.
     Second, consider that all 386 UNIX's have demand paged, virtual memory
kernels which typically give 10 megabytes to a job, WITHOUT segment
limitations, and with vastly more efficient memory management for multiple
jobs and users.
     Third, I have found that when running 386 code, as opposed to 286 code,
the performance of the INBOARD is truely excellent, exceeding the ISBC386
or the old 16 mhz COMPAQ even using 16 bit memory off the bus. When using
the 3 meg of memory on the INBOARD and daughtercard, the performance is
about 27% better. 
     Summarized results: The INBOARD does 3600 dhrystones out of standard
AT 1 wait state memory, about 4400 out of no wait state memory, such as
provided by a cheap EVEREX 3 meg ram card, and 6000 dhrystones out of
32 bit memory, which is the equivalent of a SUN 3/260, a $50k machine.
The cost of the INBOARD is typically $1000 retail, with the special
cable. You must be careful that your machine adheres to or can be made
to adhere to standard AT timing for the motherboard memory. INTEL has
a compatibility specialist to help you answer that question; ask for "Al".
Note that SCO provides specific support for the INBOARD in the form of
a boot line switch which switches the INBOARD into the high speed mode.
     I wish I could discuss my results in detail, but the editors would
be extremely annoyed if I scooped myself. Please do not mail questions
in this line, as I will regrettably be unable to answer them. I did
find, through extensive torture, that the INBOARD is an extremely
reliable device, and I use two of them for AI R&D here at Automata
Design Associates.

                       Bob Morein

> Steven P. Donegan
> Sr. Telecommunications Analyst
> Western Digital Corp.
> donegan at stanton.TCC.COM