LT spice speed

I'd export the netlist before and after moving the power supplies. If the files are exactly the same then it's hard to see why the simulation would be different.

He's right. If you moved the parts and got a change in simulation speed, you must have changed the net-list. I'd look for a typo in the name of one of the 3.3V power rails. It probably won't treat "3.3V" as if were connected to "3.3 V", but that's merely the first place to look.

I had one sim that was running picoseconds per second. I added a 1K resistor that was just grounded on one end and increased sim speed about 7 orders of magnitude.

Changing a resistor from 50 ohms to 51 can have the same effect.

LT Spice seems to be a chaotic system where tiny changes can grossly affect convergence. Moving a part will change the ordering of the net list and change the matrix. Butterfly effect.

I wonder if comments can do that too.

LT lets me set a max time step but not a min. Better yet, it should let me specify a fixed time step.

It sounds like some matrix in Spice is very close to singular. Then it can become chaotic. And dependent on the phase of the moon.

Joe

I don't think that there's any matrix inversion in Spice. It's just numerical integration of a lot of parallel differential equations.

The problem with numerical integration is that you have to keep the time step small enough that none of the differential equations - even the fastest moving outputs - can run away.

I wrote such a numerical integration program to model the gas phase chemical reaction I studied for my Ph.D. project (and extended it to cover gas flows and self-heating that accompanied the reaction). I ended up putting in tests for run-away on every step (which was recommended in the textbooks I was working from, and didn't slow the program significantly).

Of course I ended using the program to extract rate constants from experimental data, and a non-linear multiparameter least square curve fitting program adjusted the rate constant (along with the starting and equilibrium concentrations) until I had computed curves (about 80 of them) that matched my experimental data.

Spice integrates a lot more differential equations than I ever did.

Mike Engelhart talked about cutting a few corners in LTSpice tp speed it up.

John Larkin's problem may be that he doesn't seem to bother putting in parallel capacitances and equivalent series resistances in a lot of his components, and some of them may end up getting together as unintended very high frequency resonators, needing very small step sizes.

How his floating 1k resistor grounded at one end might have killed such a resonance isn't obvious - one would imagine that there'd have to have been a typo-generated invisible (and unsuspected) link for that to happen.

On 8/18/2025 6:59 AM, john larkin wrote: ...

I recall some past discussions you had with folks in this group about using alternatives to LTSpice.

That how Pspice ran 10 years ago.

It seems they updated the core, combined with Auto Convergence, I almost never see those kinds of problems any more.

Some of my guys are playing with Qspice. It was written by Mike E, same guy. I don't know if it converges better.

ECA was cool. You set the time step. It just ran and occasionally said "failed to converge" or "division by zero" and kept going.

Anything else in the sim?

Did nodes change enumeration in the process?

Were they originally drawn in or section-duped?

RL

Just this: I did poach part of the circuit from another project. Might that matter?

After this worked, I moved the 3.3 supplies and it got faster.

This ran at 100u of real time until I did many tweaks, and got it to average about 500.

Only to the extent that you can choose to hide or display some of the properties of your parts. The other project might have made choices that you weren't aware of.

You might have to compare .asc files to find out.

It was my circuit, so I doubt it.

Human memory is fallible - yours more than most - so you shouldn't be all that confident.

That might look a bit obsessive, but it's the only way to be sure.

How did you get to be such a repulsive person? Clearly you enjoy it.

I nearly responded to that with something similar. I wonder if psychology has a name for someone who thinks they know more about someone else than someone else knows about themselves. Perhaps he should ask his wife.

Failing to flatter people, even if they feel they deserve it, may count as mildly anti-social in some circles, but it never rates as repulsive.

Actually flattering egomaniacs - the sort of thing that each of Trump's executive team had to do to get the gig - is actually pretty repulsive behavior.

Do try to make your insults more plausible - if nothing else it would add a bit of class to the group (not that you'd notice that).

It all depends how much the offender actually knows. Somebody who gets close enough to the truth to needle the offended victim tends to be labelled an effective critic. The word "satire" sometimes gets deployed, but around here the tendency is more towards the "lampoon".

It's not really a category where psychologists like my wife are all that interested - it's more literary criticism.

Sadly, he seems to be pathologically envious of you, apparently looking for every opportunity he can find (or manufacture) in your posts to make an insulting comment. He really could add something of value, but his negativity discourages productive discussion far too often, at least for me.

Ed