This is a simple clock buffer with 0.001p step. It's obviously using 0.5ns at points instead.

Xyce uses an adaptive timestepping algorithm. The initial dt is varied throughout the sim. There are some flags that can change that behavior I believe.

But it is supposed to be limited by the tstep as a maximum, right?

I....don't remember that clause in the reference manual. Possibly?

Users may control the maximum internal step size by specifying the step’s ceiling value in the .TRAN command (see the Xyce Reference Guide [3])

What's your .TRAN invocation look like?

.TRAN 0.001p 70n

Ah ok. The initial and ceiling vals are different parts of the invocation

where is that from?

The xyce reference manual

Oh, I just saw all of this.

Ah, I'm looking at the user guide

You probably have this sorted out, but the time step ceiling is the 4th argument of the .TRAN line. The 3rd argument is the time at which it starts transient output, I believe. So usually you’d just have that be zero.

oh, I guess other spice are like that and I've always misunderstood

I’m not sure about all the other spices, but we patterned this after what we saw in Spice3 years ago.

Yeah, this is definitely user error!

We also added one thing to it, which is the “schedule” option. This allows different max time steps in different time ranges. That is also set on the tran line, as an expression.

As I understand it, the old

.tran <tstep> <tstop> <tstart>

comes from the early days of SPICE when the plots were all text-based. So,

tstep

is how big the time interval will be per line printed by your old needle printer on the old green-and-white paper. Ah, the bad old days when the printers went BRAAAAAP!... BROOOOP!