linux-mips-fnet
[Top] [All Lists]

MIPS Magnum 4000 history from a faq - apocraphal?

To: "alhaz@xmission.com" <alhaz@xmission.com>
Subject: MIPS Magnum 4000 history from a faq - apocraphal?
From: "Eric Jorgensen" <ericj@parlant.com>
Date: Fri, 14 Aug 98 11:05:54 +0700
Delivery-date: Fri, 14 Aug 1998 11:05:49 -0600
Envelope-to: alhaz@xmission.com
Priority: Normal
Reply-to: "Eric Jorgensen" <ericj@parlant.com>
Sender: alhaz@xmission.com
Gregory Shippen on c.s.m writes; First a little history. The secret, I
guess, regarding the Magnum was that the MIPS/SGI Magnum 4000PC/SC systems
were originally designed by engineers at Microsoft.  The design was later
purchased by MIPS Computer Systems (before their merger with SGI).  It was
chosen over an internal R4000-based workstation design that was well along
at the time.  The decision was based mostly on the fact that Microsoft had
already done the NT driver support for their design (code named Jazz) and
MIPS didn't wish to invest the resources necessary to write the necessary
HAL and drivers for their internal design if there was already an R4000
design around at Microsoft.  It also didn't help that the internal
workstation design used the DEC Turbochannel expansion bus rather than the
EISA bus used in Jazz and was something like six months later than than the
Microsoft design. 

The decision to scrap the internal design was an agonizing one for all
involved.  Until the R4000, all other MIPS processors had been brought up
on internally designed MIPS systems.  The R4000 turned out to be the first
(and last) to be initially powered on with systems designed out-of-house.
The internal workstation design was originally intended to be the bring-up
vehicle for the R4000 (the project was code named Aftershock -- after
several smaller Silicon Valley earthquakes that occured in the years before
the 1989 Loma Prieta earthquake).  Unfortunately, the design team, in
retrospect, picked the wrong expansion bus and was later to the finish line
than the lean-and-mean (and not thinking about high-volume) Microsoft team.

Once the design was purchased from Microsoft, a number of logic and
mechanical changes were made to make the design volume shippable.  (The
original design was intended to be a low-volume platform for internal code
developement only.) The only significant architectural change was to
completely redesign the audio subsystem (that was my job :-).  The only
other significant change was to tweak the ASIC set in order to support the
R4000SC with secondary cache (the original Microsoft design only supported
the R4000PC).  All other changes were minor component changes, a relayout
of the board, and significant changes to the mechanical design.

With the design rights owned by MIPS, RISC/os was ported to the machine
running in big-endian mode.  The Magnum is the only machine ever built to
my knowledge that was truly bi-endian. Indeed the two OSes supported used
different endiannesses.  The machine, on power-up, was little-endian.  If
the boot PROM was detected to be RISC/os, the machine reset itself and
switched to big-endian. If it detected a little-endian NT ARC prom, it
remained in little-endian and simply jumped to the PROM code.

Essentially, Windows NT was written on MIPS R4000 machines (no suprise here
-- this has been somewhat general knowledge).  What wasn't generally known
was that the R4000 hardware on which the code development was done was
designed by Microsoft themselves and not MIPS.  Later, after the Magnum was
shipped, Microsoft bought many of them to replace their then aging original
Jazz hardware.

Surprisingly, I believe that until NEC and later NeTPower designd MP
R4000MC systems, virtually all MIPS-based NT machines derived from this
original Microsoft design.  This included machines from MIPS, NEC, Olivetti
and Carrera. The only single-processor design which booted NT I know of not
derived from the original Microsoft design was Deskstation...



<Prev in Thread] Current Thread [Next in Thread>
  • MIPS Magnum 4000 history from a faq - apocraphal?, Eric Jorgensen <=