kds@i80386.UUCP (Ken Shoemaker) (10/31/85)
First, thanks for all the offers of congratulations! Second, an apology
for this note being a little late, I was out of town for a while, but as
a result, unix has now been running on the 386 on 3 continents!
Second, I'm sorry that the information number posted doesn't work outside
the US. If you want the info, and cannot call the number, just drop me
a note, and I will call it for you. Please do this only if you can't call,
as I have a life to get on with, too! If you've already sent me a note,
I have already called. If you don't get your stuff within a few weeks,
send another, your note probably ended up in someone's bit bucket!
There has been a bit of confusion about plugging the 386 into an AT. While
it is possible, it requires an interface card, since at 132 pins, the
386 isn't drop-in, pin compatible with the 286. What the 386 does provide
is a 16-bit bus option, which makes running the 386 into the AT socket
(or any other 286 socket) pretty simple. While the card that I have here
uses 7 chips to do it, I've heard that some of our customers get away
with as few as 3. Also, with the simple minded interface cards we use,
we introduce a wait state over the 286, and run at the same speed as the
286. This makes us incredibly bus bound, and what this means is that you don't
get much of a performance improvement with this interface card running
straight 286 code. In fact, depending on the application, yet another
wait state can lead to a slight performance degradation. Of course, if
you wanted to go all out, make an asynchronous card, and run the 386
at 16 MHz, you'd probably get more of an improvement. And, again, if
you were to re-compile your programs to take advantage of the architectural
enhancements of the 386, you could also get more of a performance improvement.
For us, the main idea of the interface card was to allow us to run as much
software as early as possible to check out the chip. Being able to plug
the 386 into the PC-ATs out there and still have a functional PC-AT (you
really still can run Lotus, Wordstar, Flight Simulator, whatever) is
a great way to check out the 386 and to allow early software development
for the 386, but it isn't a real good way to benchmark the part for
performance.
As to the operating system that we are running here, this is a modified
version of xenix 286, release 3 for the Intel 286/310 box. Using the same
interface card as is used in a PC-AT, I have plugged the 386 into the
286 socket on the 286/10 card. This is a fully protected mode 286
operating system, using the integrated memory management and protection
hardware of the 286, and now the 386. While it would be possible to
compile and execute programs on the box (in fact, it has been done),
this is not an appropriate showcase for the 386 performance, since the
compiled code would also run on a 286, and because we are running over
a 16-bit bus at 1/3 the clock rate of the 386 with (I think) 3 wait
states. I am working right now on being able to run 386 code on
a 386 system. This entails adding a loader to xenix to handle the
32-bit programs, and porting xenix 286 over to the 386/20 card I have
right here (and a few other modifications to the kernel, which could
be major or minor, depending on how carried away I get). I really don't
know when this will be done, since I am going to be away again in a few
days for a few weeks. In the interum, others here are looking for alternate
ways to run benchmarks. Their results should be available shortly.
If you are looking for other information about the 386, you can look in
the October 17th issue of Electronic Design, wherein is contained our
intro article, which is the first of a three-part series. Of course,
most any other electronic magazine, newspaper, what have you, has had
some mention about the 386. We even made the Wall Street Journal and
the London Times.
And finally a correction. Pete Kaiser at DEC notified me that the
MicroVAX II was running Ultrix on the net before they announced it.
I guess that puts us in good company!
-------------------
Ken Shoemaker, 386 design team, Intel Corp., Santa Clara, California
{hplabs,pur-ee,qantel,amdcad,dual}!intelca!i80386!kds
{hplabs,pur-ee,qantel,amdcad,dual}!intelca!kds