AT&T WGS 6386

[Bill Mayhew]

I did some of the testing in the aforementioned Unix Word article.
I would attribute much of the 6386's good showing to the fact that
AT&T's Unix and all its utilities are apparently '386 native code,
while much of SCO's Xenix is still '86 (not even '286) code.

The Olivetti herritage of the 386WGS still is quite strong, and it
still looks very much like the classic 6300 and 6300plus.  The
peripherals were high quality.  The disk was a Micropolis and the
tape unit was a Wangtec.  The unit we have just has the standard
ATT400 / CGA compatible video card, reworked to run in a standard
bus slot, rather than the monstrositiy in the 6300.  I've heard
that a few people have had some bugs with using EGA cards.  The
Micropolis drive was large enough that it didn't mount in the
standard width WGS case; it was in a sidecar unit bolted to the
right hand side of the machine.  At the time I looked the WGS over,
it was running a beta copy of Unix, but it was quite stable and I
presume it was for all intents and purposes what shipped as the
official release.

If I had to make a purchasing desicion based on the need to run
Unix, I'd be very temped to choose the WGS above the IBM model 80
because the quailty of AT&T's Unix seems to be superior to the
Unixes that are available for the model 80.  Of course, IBM is
saying that they have intent to give Unix its due, but then IBM has
been saying that for some time and it is still a largely unfuilled
promise.  I think the quality of construction in the IBM model 80
is probably superior (as is evidenced by the fact that IBM will
write a carry in service contract for $20/month), but AT&T's
accesibility via the hotline and usenet is superior, so the WGS is
still a better choice in my book.

John did the benchmarks on the WGS, so don't write me for the gory
detials.  I just commented mainly on the WGS's contruction, etc and
ran the comparisons on the mod 80.

Oh yes, I just remembered, the model 80 suffers from a bug in the
way the 80387 is handled in protected mode.  If you get a floating
point exception while the machine is doing a DMA tranfer, the
mahcine will crash.  I've tried to aggrivate the bug to occur on
the model 80 here, but it hasn't done it for me.  In a call about
something else, one of the technical support people at SCO
mentioned it to me.  This is an Intel bug, and is not particular to
IBM.  Fortunately some fairly simple glue logic between the '386
and '387 fixes the problem.  One would presume that since the '387
is a requisite condition for running AT&T's Unix on the WGS, that
AT&T designed in the necessary glue logic at the outset (or else
they don't do DMA :-)).  If good floating point performance is
important (i.e. you want to use the 387), the above is a definite
plus in the AT&T column.

I also looked at the tower version of the AT&T '386.  That is one
heck of a machine.  It's built like a tank.  There are 12 slots and
a huge power supply.  It's an fairly expensive unit, but has very
close to minicomputer performance.  The advantage is that you can
easily drop in multiport serial boards and ethernet cards, etc from
a number of companies.  The choices for the microchannel
architecture are much more limited.

--Bill
  wtm at neoucom.UUCP