Re: Rewriting SSA. Is This A Chance For GNU/Linux?

On 4/5/25 9:07 PM, The Natural Philosopher wrote:

On 05/04/2025 23:27, c186282 wrote:

On 4/5/25 3:40 PM, The Natural Philosopher wrote:

On 05/04/2025 20:22, c186282 wrote:

Analog ... it still may have certain uses, however for
   chain operations the accumulated errors WILL getcha.
   Might be a 'near' or 'finely-quantitized' sort of
   analog - 100,000 distinct, non-drifting, states that
   for some practical purposes LOOKS like traditional
   analog. So long as you don't need TOO many decimal
   points ......

  Analogue multiplication is the holy grail and can be dome using the exponential characteristics of bipolar transistors
>
https://www.analog.com/media/en/technical-documentation/data-sheets/ADL5391.pdf >
>

   Finally ... non-binary computing, eight or ten states
   per "bit". Fewer operations, fewer gates twiddling,
   better efficiency anyhow, potentially better speed. But
   doing it with anything like traditional semiconductors,
   cannot see how.

>

Non binary computing is essentially analogue computing

>
   Ummmm ... not if you can enforce clear 'guard bands' around
   each of the, say eight, distinct voltage levels. Alas, as
   stated, those 'different voltage levels' mean transistors
   aren't cleanly on or off and will burn power kind of like
   little resistors. Some all new material and approach would
   be needed. Meta-material science MIGHT someday be able to
   produce something like that.
>

It is already done in Flash RAM where more than two states of the memory capacitors are possible
>
Massive arrays of non linear analogue circuits for modelling things like the Navier Stokes equations would be possible: Probably make a better stab at climate modelling then the existing shit.

>
   Again with analog, it's the sensitivity to especially
   temperature conditions that add errors in.

 Not really, That was mostly sorted years ago.

   Ummm ... I'm gonna kinda have to disagree.
   There are several factors that lead to errors in
   analog electronics - simple temperature being
   the worst.

   Keep
   carrying those errors through several stages and soon
   all you have is error, pretending to be The Solution.

 So no different from floating point based current climate models, then...

   Digital FP *can* be done to almost arbitrary precision.
   If you're running, say, a climate or 'dark energy' model
   then you use a LOT of precision.

   Again, perhaps some meta-material that's NOT sensitive
   to what typically throws-off analog electronics MIGHT
   be made.
>
   I'm trying to visualize what it would take to make
   an all-analog version of, say, a payroll spreadsheet :-)
>

An awful lot of op-amps.

   To say the least :-)
   CAN be done, but is it WORTH it ???
   But, I suppose, a whole-budget CAN be viewed
   as an analog equation IF you try hard enough.

The thing is that analogue computers were useful for system analysis years before digital stuff came along. You could examine a dynamic system and see if it was stable or not.

   Well, *how* stable it is ........
   Digital is always right-on.
   So what do you NEED most - speed or accuracy ?

If not you did it another way. People who dribble on about 'climate tipping points'have no clue really as to how real life complex analogue systems work.

   I'm just gonna say that "climate" is beyond ANY kind
   of models - analog OR digital. TOO many butterflies.

   Now discrete use of analog as, as you suggested, doing
   multiplication/division/logs initiated and read by
   digital ... ?
>

Its being thought about.

   And we shall see ... advantage, or not ?
   Maybe, horrors, "depends" .....
   The "real world" acts as a very complex analog
   equation - until you get down to quantum levels.
   HOW the hell to best DEAL with that ???

   Oh well, we're out in sci-fi land with most of this ...
   may as well talk about using giant evil brains in
   jars as computers  :-)
>

Well no, we are not.
Digital traded speed for precision.

   I'd say digital traded precision for speed ...

   As some here have mentioned, we may be closer to the
   limits of computer power that we'd like to think.
   Today's big trick is parallelization, but only some
   kinds of problems can be modeled that way.
>
   Saw an article the other day about using some kind
   of disulfide for de-facto transistors, but did not
   get the impression that they'd be fast. I think
   temperature resistance was the main thrust - industrial
   apps, Venus landers and such.
>

I think I saw that too..
 Massive parallelisation will definitely do *some* things faster.

   Agreed ... but not EVERYTHING.
   Sometimes there's just no substitute for clock
   speed and high-speed mem access.

Think 4096 core GPU processors...I think that's the way it will happen, decline of the general purpose CPU and emergence of specific chips tailored to specific tasks. Its already happening to an extend with on chip everything....

   I kinda understand. However that whole chip chain
   will likely need to be fully, by design, integrated.
   This is NOT so easy with multiple manufacturers.
   I've avoided investments in higher-tech/specific-
   tech stuff. NVIDIA seemed good - but the current
   trade war is gonna put a BIG strain on them. It's
   all too volatile, too vulnerable to even small
   politics - govt or industry.
   DO wish I'd bought RHEL back near the IPO, but,
   well ..... ya never know unless you're an
   "insider".
   ANYway ... final observation ... it keeps looking
   like we're far closer to the END of increasing
   computer power than the beginning. WHAT we want
   computed is kinda a dynamic equation, but OVERALL
   we're kinda near The End.
   THEN what ?