[sdiy] Analog Modeling, with a computer!

[David Cornutt]

On Sep 14, 2005, at 2:52 PM, Richard Wentk wrote:

>
> I have no idea how much computer power is needed to do this properly 
> in real time. But I'd guess we're about ten years away.
>
>

Doing some back-of-the-envelope stuff: I recall that some circuit sims 
we used to run at Gould would take about
two days to run.  These sims had thousands of components, but only the 
most critical parts of the circuit were
modeled in the analog domain (otherwise, the damn things would still be 
running today!).
This was on a Sun-3/260 in 1986.  If we take this to be 50 hours of 
runtime, and apply Moore's
Law (dividing the run time by 2 for every 18 months elapsed since 
1986), we come up with a run time for
that sim today of about 30 seconds on a high-end workstation.

If the amount of time being simulated was a few milliseconds, the 
runtime needs to be decreased by a factor
of 2^14 to get it down to being able to run in real time.  If we assume 
that Moore's Law continues to hold,
that will be about 20 more years.  If I keep myself in good health, I 
should live to see it.