> On Sep 29, 2026, at 2:12 PM, Antonio Carlini via cctalk
> <[email protected]> wrote:
>
> On 29/09/2026 18:22, ben via cctalk wrote:
>> On 2026-09-28 4:45 p.m., Jim Davis via cctalk wrote:
>>> It's better than it's 2+2=5 (for large values of 2) and reporting the
>>> moon is made of green cheese days.
>>> last week, In a mildly pissed session,
>>> We/i/it did a sketch of an analog tensor cell, just op amps and floating
>>> gates
>>> Digital on the in and out, but a 1M^2 cells in the core. AI's agree,
>>> It looks good.
>>> Give it a try, Promise you won't get your hand blown off playing with
>>> improvised explosives.
>>
>> And how soon will ****** start with hidden charges for AI service?
>> Normal google searches seem just skim off top five sites and repeat from
>> them.
>> What if I want to find all the computers made in Eastern Europe between 1958
>> and 1975 18 to 30 bits wide, and the data translated to English.
>> When a AI can do that let me know.
>
>
> Oddly enough I've been compiling a list of non-power-of-two machines and Ai
> does a good (or at least credible) job of turning web pages I don't
> understand into ones I think mostly match DeepL and Google Translate.
>
> Here's a starting list. I've not checked it but a few of the machine son
> there (like some of the Odra ones) are real (in the sense that I've
> independently found info about them, not that I've ever seen them ... but
> then I've never seen a PDP-1 either).
>
> Now I have to decide whether the Dnepr-1 counts as 26 bits for my list ...
> should parity/sign count?? Oh, and I was missing a 29-bit example ... so more
> checking to do (you never know, it might have imagined it!)
That's certainly a start but just the bare beginning. I suspect that, unti
lthe IBM/360, most machines had word lengths not a power of two. For example,
none of the DEC machines before the PDP-11. Older IBM machines had assorted
word sizes; CDC and Burroughs ditto. I worked on a 1970s era Philips machine
with 24 bits, and there were other minicomputer with that word length
(Modcomp?). Electrologica had 27 bit words. The MC research machines had word
lengths of 30 and 34 bits (the 34 bit one, ARMAC, was a successor of the 30 bit
ARRA 2 with a very similar instruction set, two instructions per word, but the
address space grew from 1k to 4k so the word was made 4 bits larger). ENIAC
was 40 bits I think.
I would count the sign since it's part of the data representation, but not the
parity bit. And on Burroughs mainframes the 3 tag bits are usually not
mentioned; if you do include them (sensible, since they carry semantic
information) you end up with a word size of 51 bits. Not often you run into
prime numbers for the word size.
paul
