In article <1993Jun16.225356.9610@amscons.com> bob@amscons.com (Bob Amstadt) writes:

quoted text
>   I don't want to discourage you, but there are some problems to consider.
>   If you create the fastest CPU board in the world, you will soon find
>   that you spend all your time waiting for I/O (most notably disk).
VALID !!!

quoted text
>   I'm not saying this is a hopeless cause, instead I would suggest including
>   the hard disk interface on the CPU board.

Need not be on the same bus, as bandwidth of cpu bus should be around
300 MByte for a R3000 50 MHz, and most disks deliver at less than
10MBytes/sec, some IPI-drives at 45 MBytes/sec.

quoted text
>   Many high-performance CPU's
>   made for passive backplanes take this approach.  If you choose SCSI/SCSI-2,
>   there are a wide variety of chips that would do most of the work for you.

On a fast system, you would prefer to have I/O handled independently
of the CPU. Ever considered how expensive an interrupt service is on a
R[34]000 running any kind of non-realtime operating system?

On pre--Indigo SGI there were just interesting solutions to this very
problem...

Put in an additional 20US$--microcontroller ....

quoted text
>   Also, consider including a daughter board connector on your CPU card.
>   This will give you high-performance expandibility.

NO because of RFI and buffering. MOS--Microprocessors tend to drive a
rather small capacitive load. To my best knowledge, only alpha allows
you to use ECL--interface, which would make your proposal rather
useful.

BTW: Control Data used the R6000 in ECL ....

quoted text
>   I don't know if
>   you would ever use it, but if your going to take on this project, don't
>   build something that is already obsolete.

Full heartly agreed. I'd prefer a R4000 to a R3000.

BUT there are two problems remaining:

        As MIPS was bought by SGI, the R[34]000 became more or
        less a property architecture. And Silicon Graphics Inc. is
        able to supress competition by price of their licence and
        processors. Up to now no larger clone maker has decided to build 
        systems based on MIPS--processors, they prefer th SPARC/SUN--market..


        Because of the to--be-expected low production volume, it
        might become a RATHER expensive experience ....

best, adams