Perkin-Elmer 3205?

[jbray at bbn-unix]

From:  James Bray <jbray at bbn-unix>

The Perkin-Elmers are indeed fast, with the synchronous machines (3210,20,30)
cycling at 200ns and doing register-register adds in 450ns, and the async
machines (3240,50) being somewhat faster. I don't know anything about the 3205.
It is not quite correct to say that they don't have virtual memory: the 
machines with the MAT (memory address translator) which obsoleted the MAC
(which I don't think they make anymore) have all the necessary hardware, as
the MAT supports something resembling two levels of paging/segmentation, with
hardware dirty-bits and the whole thing. OS/32, however, and probably also
their unix, don't support paging or virtual memory: that is, they just don't
use the stuff. Inside sources told me that they had thought about it for a
long time, and decided that with a 16-meg address space on their larger machines
and 4meg on the smaller and memory getting cheap, they just didn't need it. 
Remember that the major target of their stuff is people doing Fortran number-
crunching and realtime-type stuff: for these applications, one wants physical
memory, not virtual. Their stuff is not really built for timesharing.
  But the hardware support is all there; it just needs to be hacked into an OS.

--Jim Bray