29 May
2011
29 May
'11
7:20 a.m.
For the same reason anyone implements any other CPU in
an FPGA -- so
you can use that CPU architecture in a modern design or as a hardware
simulation of a vintage design.
If a hobbyist wants to play with Transputers, the easiest thing to do
is just wait until some become available. The things have been
floating to the surface for some time now, probably because of
hoarding.
I think there's a difference between commercial
success and not
working well. ?"Not working well" implies to me that it had some
functional deficiency. ?Transputers ended up not being commercially
successful because they were expensive and not many people wanted to
program in Occam, despite whatever advantages CSP may have had.
Inmos managed to miss both of those targets, although I am hesitant to
call it a complete commercial failure. They managed to sell a good few
of them, but not the "take over the world" amount they certainly were
boasting about. The simple systems worked fairly well, due to the raw
speed of the processors, but the massive systems stumbled because of
the networking. Things basically did not scale well.
The Transputer story reflects the whole MPP industry of the 1980s -
lots of neat ideas, and lots of "close, but no cigars". Meiko, Convex
Exemplar, Thinking Machines, Maspar. Lots of what we learned falls
into "what not to do" territory.
--
Will, sacred cow tipper