Is there a way to force ngspice to use 4 cores during simulations, currently it uses only one when I invoke it from Xschem?

You need to create a .spiceinit file with this in it: set num_threads=4

Thank you, that worked. It uses all 4 cores but the utilization is ~40% per core for the first three cores and ~90% of the 4th. This is when I set it for 4 threads (so it has improved two times roughly). It is ~40% per core for all 4 cores when I set it to 8 threads (as viewed by htop). Is that expected, why is it not utilizing 100%?

Parallelizing code is hard! 🙂 The things that can be parallelized are: 1. BSIM model evalution 2. Solving the linear system between each time steep #1 scales well if you have a lot of devices, but you will then be bottlenecked by #2. #2 depends on how interconnected your circuit is whether it can be divided into parallel problems. I'm not sure if/how well ngpisce does this but Xyce does a really good job.

yes I have considered using Xyce but after seeing this results https://github.com/mabrains/xyce_sky130/blob/main/testcases/xyce_vs_ngspice/comparison_result_ac.csv (207.693% error or rather deviation from ngspice results) I stuck with ngspice because of better accuracy

Do you know if ngspice is accurate?

There are some good recommendation on running ngspice with skywater here http://ngspice.sourceforge.net/applic.html