LispOS on a 68000

Mark J. Dulcey
Wed, 20 May 1998 23:52:31 -0400

Will Hartung wrote:
> The Alpha was discussed because of its 64-Bit architecture helping
> make a tagged less expensive. Little bits of muttering about the Sparc
> having some instructions to support tagged structures.

The other nice thing about Alpha is that there are inexpensive
surplus machines and motherboards out there, and even some
not-too-expensive motherboards based on the new 21164PC chip. The
original Alpha chip lacked byte-handling instructions, so it
might not be so good for handling tags; the 21164 and 21264 are
better bets. On the other hand, 21064 and 21066 systems, such as
the popular PCI33 motherboards and UDB systems, are VERY cheap.

There isn't so much in the way of cheap SPARCs out there, and
nothing with an UltraSPARC chip (the new 64-bit version) is going
to be inexpensive; if you really need an UltraSPARC system that
isn't too horribly expensive, the best bet is probably to buy an
Ultra 5 (about $4K new; a bit pricey to pick up just to play
> And what's particularly bad about the Intel architecture WRT a Lisp OS
> built from the metal up. And are the problems with the Intel bad
> enough to warrant giving up on essentially free motherboards with
> "proven" designs.

The main shortcoming of the x86 architecture is its paucity of
registers. That makes it harder to squeeze performance out of a
system, because it tends to be badly memory-bound. An additional
problem is that its lack of regularity (lots of special-purpose
registers and so forth) makes it hard for compilers to produce
really good code.

Right now the primary RISC architectures (Alpha, UltraSPARC,
PowerPC) have at least a 2-1 raw performance advantage over x86
(with good optimizing compilers and WITHOUT hand-optimization;
good hand-coding can narrow the disadvantage of x86), and it is
likely to continue to widen.

Mark J. Dulcey     
Visit my home page:
Visit my house's home page: