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

On 4/5/25 9:13 PM, The Natural Philosopher wrote:

On 05/04/2025 23:34, c186282 wrote:

On 4/5/25 4:00 PM, Farley Flud wrote:

On Sat, 5 Apr 2025 15:22:23 -0400, c186282 wrote:
>

>
    Digital ... note that clock speeds haven't really risen in
    a LONG time. They can, to a point, make them 'work smarter'
    but how much more ? Not all tasks/problems lend themselves
    to parallel processing methods either.
>
    So, yea, we're pretty much there.
>

>
The supercomputer people would disagree.
>
Supercomputers, based on Linux, just keep on getting faster.
>
The metric is matrix multiplication, a classic problem in cache
management.
>
I don't know about the architecture of supercomputers but
the limit seems to be still quite open.

>
   Matrix mult is a kind of parallelization ... and we
   still have some room there. But not every problem is
   easily, or at all, suited for spreading over 1000
   processors.
>
   https://en.wikipedia.org/wiki/Amdahl%27s_law
>
   Super-computers can use exotic tech, at a super price,
   if they want - including superconductors and quantum
   elements. NOT coming to a desktop near you anytime
   soon alas ......

 Well the main use of supercomputers is running vast mathematical models to make sketchy assumptions and crude parametrisations look much more betterer than they actually are..

   Even 80s super-computers made it unnecessary to TEST
   nuclear weapon designs - the entire physics could be
   calculated, a 'virtual' bomb, and RELIED on. Even Iran
   can do all that and more now.

Real racing car and aircraft design uses wind tunnels. CFD can't do the job.

   Um, yea ... really COULD be entirely virtualized.
   ACCESS to such calx capabilities still isn't there
   or affordable to ALL however. Do you think that
   AirBus/Boeing/Lockheed build a zillion wind-tunnel
   models these days ? Likely NONE. The airflows, the
   structural components ... all SIMS.
   WHAT can be done with "quantum" is not entirely
   clear. Again it's not best suited for EVERYTHING.
   The #1 issue is still the ERROR rates. As per QM,
   where things can randomly change Just Because,
   these errors are gonna be HARD to get around.
   STILL no great solutions. Got a design for a
   "Heisenberg compensator" ??? If so, GET RICH !!!
   There MAY be some pattern in the QM errors that
   can be matched/negated by some parallel QM
   process/equation. We shall see.