25th Anniversary of the Personal Computer" 4196
I dunno. Lisp is, I suppose, in some ways, easier to compile than Fortran (particularly if you get to write the compiler in lisp!). Clbuttic AI programs of the 1970s didn't typically rely on enormous databases. They all fit on the '10 one way or another. A lisp program can load and unload functions as it goes along; I suppose the point under consideration might simply be how much stuff you would need in memory at one time. Not to forget also that lisp storage is all (implicitly) dynamically allocated, and reclaimed by the garbage collector when it is not longer referenced. Did I mention already: I dunno.
25th Anniversary of the Personal Computer" 4202
On Tue, 15 Aug 2006 09:41:28 +0200 Perhaps - OTOH I found MP-M systems with a 6MHz Z80 and 256K of...
The quote is correct. There is nothing about the context that I can see that would cause the quote as extracted to be misleading. (By-the-bye, Barb: How are your plans progressing to get yourself set up with a real computer so that you can check these sorts of things for yourself from home?)
Think of the quote in the context of the times. Researchers felt that they were already getting close to usable intelligent systems with natural-language front ends within the confines of the PDP-10 address space. Increasing resources by an order of magnitude surely must have seemed like it ought to do the trick. Additionally, the authors were likely thinking of natural-language front ends that needed to understand only a well-defined, severely constrained micro-world of some sort. (Winograd's SHRDLU was still a rather recent succĆs fou at the time the lispm was conceived.)
The memo reports on a Lisp Machine design with a 24-bit virtual address space (and 32-bit words). Maclisp (and, for that matter, also Interlisp) on the PDP-10 at the time effectively used an 18-bit address space (a cons being stored as two pointers in one machine word -- the PDP-6-PDP-10 36-bit architecture was designed with this use in mind, as I imagine you know).
So, if I'm doing the sums right, 64 times the address space of the '10, but it takes 2 words to store a cons, so let's say 32 times as many cons cells as conveniently fit on the '10 (256k conses on the 10, 8M conses on the lispm). These are of course just back-of-the envelope estimates that conveniently ignore the fact that the whole address space isn't given over to conses (and maybe some real lisp-implementation maven will be along presently to point out any serious errors I may have made) -- but the general picture is still clear: this architecture represented an increase of more than an order of magnitude in the resources available to an A.I. researcher, even before you take into account the benefits of not having to share the machine. (Note that this, and earlier and contemporaneous work at PARC, is essentially the birth of the Dedicated-Workstation-As-We-Know- It -- or rather, as-we-know-it modulo the exotic architecture.)
25th Anniversary of the Personal Computer" 4197
needed. Yea, well. It was a VAX attack. That's what they called it? Think about it. There was many millions spent to emulate what a gal...
A relevant quote (MIT AI Memo 444, page 3), which perhaps supplies some further context for you:
"Another problem with the PDP-10 LISP implementation is the small address space of the PDP-10 processor. Many Lisp systems, such as MACSYMA and Woods's LUNAR program, have difficulty running in an 18-bit address space."
Anyway, returning to the nominal subject of this thread, do we not all feel somehow cheated that the CONS machine and its progeny lost and the x86 won the processor wars? (Especially that the x86 won.) I'm trying to run my life on a glorified calculator chip, for goshsakes!
-- Roland Hutchinson╩╩╩╩╩╩╩╩╩╩╩╩╩╩Will╩play╩viola╩da╩gamba╩for╩food.
25th Anniversary of the Personal Computer" 4198
the I Right. Now, please listen. VMS was not this simple. It should have been but the pricing nazis created a dozen CATCH-22s with each product. The developers didn't help either because there wasn't...
NB mail to my.spamtrap at verizon.net is heavily filtered to remove spam.╩╩If╩your╩message╩looks╩like╩spam╩I╩may╩not╩see╩it.