Re: Remember "Bit-Slice" Chips ?

On 12/12/24 4:35 AM, D wrote:

  On Wed, 11 Dec 2024, 186282@ud0s4.net wrote:
 

On 12/11/24 5:08 AM, D wrote:

>
>
On Wed, 11 Dec 2024, 186282@ud0s4.net wrote:
>

On 12/10/24 8:38 AM, D wrote:

>
>
On Tue, 10 Dec 2024, 186282@ud0s4.net wrote:
>

On 12/9/24 8:25 PM, rbowman wrote:

On Mon, 9 Dec 2024 13:58:47 -0800, John Ames wrote:
>

Any chance of this conversation returning to anything even *slightly*
more relevant to *nix, computers in general, or, like, *anything* else?
Maybe I should dig up some old Francis E. Dec rant for a more coherent/
relevant refresher course...

>
Feel free to start a thread.

>
>
 Ummmmm ... I just TRIED with the "Bit-Slice" topic.
 Jumped IMMEDIATELY back to 'non-OS/Computer stuff'  :-)
>
 Was HOPING for discussion/insight into 'alternative'
 schemes for 'CPU's and such derived from older solutions.
>
 Houston, we have a problem .............
>

>
Your thread was perhaps not interesting enough? Try again! ;)

>
>
 Awwww .... OUGHT to be an interesting topic, especially
 as we're bumping up against Moore's end-point. One or
 two more gens and we're literally at the atomic scale ;
 where to go from there ?

>
Quantum computing of course! Otherwise, we'll just continue to scale out I assume.

>
 "Quantum" has NOT been going well. People are finding
 ways to REDUCE the error rate, but it's still too much.
 Cyro is also kinda required.

 The thing that puts me off quantum is that media loves to hype it. That to me, is a sign that it is nowhere near being ready for anything productive. But I am not a physicist! But based on this group it does seem I am more right than wrong.

   Google just blew its horn today about it's new quantum
   chip - solved some 1000x-age-of-the-universe math
   problem in about five minutes.
   https://blog.google/technology/research/google-willow-quantum-chip/
   BUT, that particular problem was especially easy
   to address with quantum methods ....

Is there an established Moores end point?

>
 I'm gonna say it's "at the few atoms" zone. However
 you still need to make room for the connecting leads.
 Gotta be able to FAB such things too ... and you're
 well into the X-ray zone there.
>
 So ... I think we're approaching A Problem here.
>
 The kind/meaning of 'computing' has kinda shifted
 recently due to 'AI' - Nvidia rules there - but
 the chips are faster at "AI" sorts of stuff, not
 general/all-purpose.
>
>

I would imagine once we hit that end point in terms of regular cpus, the only direction left would be purpose built cpus on other technologies for niche use cases such as biological computing, quantum computing, optical etc.

>
 Bio is gonna be too SLOW. Quantum, we've discussed that.

 Isn't the idea behind bio massiev parallelism? So yes, the computation might be slow, but if you have millions of molecules performing it in parallel you do get fantastic results if the problem you are trying to solve fits the nature of bio computing?

   But again ... NOT all problems are especially well
   solved with massive parallelism any more than all
   problems can be Q-computed worth a damn.
   So we're back to the more modern question of what
   "computing" MEANS. All was clear with UNIVAC, but
   since then ...

 Pure photonic - including some rough analog of photonic
 transistors ... MAYbe. I keep hearing bits of news which
 suggest those MIGHT be practical someday. Still, don't
 see them being THAT much faster - the S-o-L in crystals
 and fiber and such is a limiting factor. Who'd have ever
 imagined the S-o-L would be TOO SLOW eh ? Indeed it's
 already a communications pain in the ass.
>
 Photonic switching elements don't switch instantly
 either ... MAYbe some different def/tech that's not
 really so much 'switching' per-se ? Interference
 patterns ?
>
 Now something that will properly support, say, deca-state
 logic ... ? Transistors don't do that well, but smart
 photonic design, perhaps. A lot more 'getting it done'
 per gigahertz  :-)

 Seems like photonic is the winner for the moment.

   Don't see any other direction. We're already kinda
   bumping-up against Moore even now with conventional
   electronics. For anything needing linear calx, I think
   we MIGHT get a 10x improvement and that's IT forever
   with transistor-like electronics.
   USED to take "Photonics Spectra", aka "Photonics"
   mag back in the day. MANY promises - but VERY VERY
   slow realization. Is still published :
   https://www.photonics.com/
   Useful photonic 'circuits' are NOT easy to realize.
   Some of the new "AI" stuff MIGHT make some of that
   easier by looking at a zillion solutions/variants
   a second and finding exploitable patterns.
   Again, 'circuits' as we think of them may not be
   the best way to use photonic tech - still have
   this weird vision of using interference patterns
   rather then ph "transistors". Light has some odd
   properties that are easier to get at than by
   using electrons.

Could regular cpu:s get some extended life by a change of materials or some other tweaks to the current design?

>
 Could ... but instead they'll make new chips.
>
 Oh, news today ... if you have an AMD box DO look into
 the "BadRAM" exploit - a sneaky back-door way for Vlad
 to spy on your 'protected' data.
>

 Better innovate SOMETHING, otherwise we're gonna see
 'peak computing' when it's become clear we need thousands
 of times that for the Really Cool Stuff.
>

>
>