On Wednesday 18 August 2010, Oliver King-Smith wrote: > I am under the probably incorrect impression that LtSpice is > actually a better than ngspice and gnucap. What do you > think the benefits are gnucap vs Ltspice?
Better for what? You got that impression because somebody is promoting it and you believe the ads. The biggest benefit of LTspice is that people like the user interface. Also, it comes with a library of stuff that some users like. The biggest drawback of the geda/gnucap combo is that the interface between gschem and gnucap works very badly, hence my request for help. Another issue with gnucap now is that there is a big difference between the stable branch and the development branch. Some of the most important features are only in the development branch, which is not the one you get with most Linux distros. For the benchmarks I have run, Gnucap outperforms LTspice and NGspice for medium to large circuits, sometimes by a huge amount. I recall one where Gnucap completed a transient analysis in a few minutes, NGspice took about 8 hours, and I gave up waiting for LTspice, which had produced no output after running it overnight. Also, Gnucap produces output along the way, so even with a slow run you get to see something soon, but either Spice doesn't show the user anything until it is all done. This is not a random difference. I fully understand the reasons for it. Another benchmark, run a long time ago, comparing a predecessor of Gnucap to an expensive commercial simulator that specializes in power grid analysis, the predecessor of gnucap outperformed the commercial package for power grid analysis. Gnucap is a lot more flexible that any spice. It has plugins for lots of stuff, including models, measurements, simulation languages. You can add new (lots of stuff) without recompiling or reinstalling. It takes several simulation languages, including spice, spectre and Verilog. Gnucap has more flexibility in how you run it, for example, you can do an AC analysis at an instant in time, such as a snapshot from a transient analysis. Spice only lets you do AC at quiescent. To see an example of where this matters, try doing an AC analysis of a class B amplifier. Gnucap's step control works better. One example of where this shows is in simulating oscillators. Gnucap is accurate enough to make distortion measurements on a sine wave oscillator. Spice isn't. Gnucap is accurate enough to properly simulate a negative resistance oscillator with a switch, and gives a correct waveform and correct oscillation frequency. Either spice gives nonsense on this circuit. Gnucap is not Spice. (in the same sense that Gnu's not Unix, or that C++ is not Fortran) Gnucap development is more focused on moving forward than bug-for-bug compatibility with legacy programs. That's just a little. But really, in free/open-source, or anywhere, you should expect tradeoffs. One will be better in some ways, another better in other ways. The question should not be which is better, but how do we make ours better. That's both better than it is, and better than others. And remember, things get better when you and we work to make them better, and worse when you see a perceived deficiency and run the other way, and worse when we deny the deficiencies and keep the status quo. The Gnucap development team is working on features for advanced users, with the intent of eventually fully supporting Verilog- AMS, through partnerships with some other GPL'd projects. What seems to be missing is the partnership with schematic and layout. It is in this area that help is most desperately needed. Gnucap/gEDA can be made to be better that LTspice (and others too) in every way, if we choose to do so. Let's do it. _______________________________________________ geda-user mailing list geda-user@moria.seul.org http://www.seul.org/cgi-bin/mailman/listinfo/geda-user