> A far more stable approach to increasing computing power is using the 
> motherboard as simply a backplane bus, with a slow stable CPU simply 
> managing the bus. Most new add-on cards are ready for this re-invented s  
> ystem design (a throwback to the mainframe). Video cards, sound cards, 
> controller cards, etc. all have their own Bios and CPU and command 
> interface and system memory access.

the Big Concept I had for computer architecture bears some resemblance to
this; but I don't know enough about the Electrical Engineering details to
know what's possible, and what isn't.

as I understand SGI's O2 architechture; there's a memory manager which
controls access to main memory; and other devices just sort of hang off it
-- CPU, Graphics Processing Unit, display processor, I/O boards, etc.
the GPU shares memory space with the CPU; so there's no "out to memory, over
the bridge, into graphics memory, into the GPU" chain; it's just "put
something in memory and tell the GPU where it is".

if this could be sufficiently generalized (and I think HyperConnect [?] and
a few other new technologies are headed in this direction); you just have a
memory controller, and 'channels' hung off it. want more number-crunching
speed? put another CPU on a channel. want better graphics performance? put
more GPUs on; and since they all share the same memory space, it should be
easy to Scan Line Interleave like the old Voodoo cards -- but may be
possible between completely different cards, instead of just cards built for
it.

to a degree, SGI's Origin 3000 architecture is like this... too bad it's out
of the price range of ordinary mortals. I hope this sort of idea works down
the price scale tho; it's just too cool to let it go the way of the
VAXCluster and NeXT...

Carl Soderstrom
-- 
Network Engineer
Real-Time Enterprises
(952) 943-8700