John Larkin said:
Spice is all double floats, no? 1e6 isn't a very big number. A current
sampler is open-loop so won't have convergence problems.
What the numbers are doesn't matter, the solver uses iterative algorithms
that save inverting the whole N x N matrix. Matrix operations are O(N^3),
so a modest circuit with 100 nodes takes about 10^6 more operations than a
basic battery and lightbulb circuit. And that's if it guessed the correct
timestep the first time, which very rarely occurs. SPICE throws away a
*lot* of computation.
Obviously, you've never simulated anything of significance, or you'd know
this. The degree of these approximations is controlled by the simulation
parameters, which experienced users understand must be optimized for a
given model to avoid slowing to a crawl (with default settings, LTSpice
likes to wander off into the picoseconds, eating glue) or generating
errors (most often "timestep too small").
Tim