LTspice 24.1 Much Slower than 24.0: "Initializing Circuit Matrix" Takes Significant Time in Former

Hi Friedrich ,

I see the difference in the simulation time.

but its not as long as 20s for me, its only about 10s-14s tops.

Nonetheless, have you tried checking for LTspice software updates yet? This could have been addressed already.

If you had already updated your LTspice, may I ask if there are changes in the expected performance of your simulation or only the simulation time had changed?

Best Regards,
EArl

Hi Friedrich , EarlJohn ,

LTspice 24.1.1 is the latest publicly available. Took me 2s in LTspice 24.0.12; 16s in 24.1.2 (soon to be released). The time difference is interesting, mostly spent in the "Initializing Circuit Matrix" message, as stated. For fun, I netlisted from 24.0 and ran the netlist in 24.1.2 and had the same result. Looks like your Gear method took the longest. I matched everything: Modified Trap.

mike

Dear Friedrich ,

Thanks for providing this test case!

The additional delay is caused by a new feature of LTspice 24.1. It compiles all math expressions and certain other parts of the simulation into highly optimized machine code. This speeds up behavioral sources by a factor of 2. But it also takes time ...

If this turns out to be problem, we need to add an option to turn this off.

Best Regards,
Mathias

Dear Mathias!

I think that your new optimization is a nice feature for ltspice

The problem is that if simulation time is small relative to optimization time, there is no benefit at all. This is true especially for the .op analysis where the simulation time is very small (1s) and the matrix optimization is as long as for the transient analysis about 10-20s !!

Therefore i would suggest that you add the option to turn this behaviour off to let the user decide to use this new feature in a sensful way

For long simulation times your optimization is a good feature (see below)

Best regards,

Friedrich

NOTE: I`ve tested my circuit with a 1000x longer transient endtime and
only one waveform variable to plot and i get 176sec for 24.0.12 and 92sec for the new optimized 24.1.1! This is impressing and shows that for longer simulations and restriced number of waveforms to plot the optimized version is much better  !! Thank you