PDP-11 = Pentium Killer?

From: Roger Ivie <IVIE_at_cc.usu.edu>
Date: Wed Apr 22 17:31:14 1998

> What would happen if we made a PC-sized PDP-11 processor using the Alpha
> technology? (On a single chip, clock it at ~300-400 MHz)
> As the PDP-11 instruction set is MUCH better than 80x86, would it outrun
> a PC? Could this be a Pentium Killer?

I doubt it could be a Pentium killer, but I could be wrong. The RISC
machines get performance by making it difficult to go to memory; on the
PDP-11, it's much to easy to go to memory.

Take, for example, the PDP-8. The worst-case instruction on the PDP-8
could require as many as five memory accesses (hmm; I forget which one
took five. I only count four for ISZ I Z 10 (fetch 10, write incremented
value, fetch from incremented value, write incremented value pointed to by
incremented value) and three for JSR I Z 10 (fetch 10, write incremented
value, stash return address at location pointed to by incremented value),
so I could be wrong) each of which depend on the previous one. You're not
going to get hot performance out of that unless you decide that the main
memory can be built using a 5-port register file on the chip.

I've occasionally wondered about doing a tight hand-coded PDP-11 emulator that
fits in the primary cache of an Alpha. If possible, you'd be using the Alpha
essentially as a programmable microengine and programming it to be PDP-11.
The reason to fit it in the primary cache is because of how the Alpha boots;
at reset, it loads its primary cache from an external serial ROM and begins
executing it. If you could fit the emulator in the primary cache, you could
think of the Alpha+SROM as a PDP-11 microprocessor.

Roger Ivie
ivie_at_cc.usu.edu
Received on Wed Apr 22 1998 - 17:31:14 BST