I'd like to hear from users of any program, but I've only used SKM. My project has about 1200 buses and 3500 components and it is just taking forever to run an arc flash analysis (like 5-10 mins actually). It looks like the program freezes, it grows faint and the titlebar says "(Not Responding)" at the end. I'm wondering if it's my computer, or the software or just the project size, or maybe all 3.

My CPU never goes above 25% when running a calculation, I feel like it should be going to 100% instead of "freezing" or taking forever. Memory stays pretty flat at 25% too. I moved all the LIB and BIN files off the network, onto my PC which helped a little. Same with the project files. I've read that SKM is still 32-bit, maybe if they made it 64-bit it would faster?

Has anybody had more success with one of the other programs? Or have any good hacks for SKM?

I suggest calling their tech support. They should connect and look at your computer and setup to get you an answer.
I work for Milsoft. Our WindMil software takes about 20 seconds to run that size system if it’s radial.

You have to remember the Z bus that SKM uses is # busses by # busses. So if you have a model that's 10 busses, the Z matrix is 10 x 10. If you double that to 20 busses, it's 20 x 20 or 4 times the number of calculations. When you start talking about 1400 busses, I don't care how fast your computer is, a 1400 x 1400 matrix will take some time to calculate (that's 19,600 * the calculations as for a 10 bus system). If you have a system that large, you most likely have alternate scenarios also. So when you run an arc flash calculation, it has to run the SC calcs and arc flash calcs for each scenario. If you have 10 scenarios, you're running those calcs 10 times, so multiply the calculation runtime by 10. And if you're using the unbalanced fault calcs, those take longer to calculate than the balanced fault calcs. All those options add up. I'd say given the size of your system, be happy that it's knocked out in 10 minutes and not 2 hours.

I'm assuming you have plenty of RAM so that it doesn't have to go to virtual memory. I've had some fairly large systems above 500 busses. What I've found I have to do is use the library on my local hard-drive and copy the model to my hard-drive also so that it's not reading the impedance data across the network and writing the results for every bus across the network. Of course you load it back up to the network at the end of each day. Also, re-boot your system daily, and before calculating a second time that day.

As Jeff S said, the larger and more complicated the system, the more scenarios you run, the more levels upstream you check for miscoordination all impact the length of time it takes SKM (or any other software) to run.

I had one project with a very large and very complicated system that would take literally 8 hours to run. I think it had right under 2000 busses, an untold number of components, at least 20 scenarios, and we decided to check up two levels for miscoordinaiton. Memory and computing power weren't an issue, and the computer used didn't appear to have any noticeable impact on the time to complete the study. I would run the arc flash before leaving for work, then come back in the morning hoping there wasn't some type of failure. And when you found that one component out of tens of thousands that needed an edit... not fun times.

As far as different software packages go...
SKM, ETAP, and Easypower all use roughly the same calculation method, utilizing Z or Y matrices. Some have optimization subroutines, but the basic math is the same, only differing in how the impedance data is stored and libraried.

The previously mentioned Milsoft Windmil uses a bus-by-bus calculation method that gives it that quicker speed. I believe it can only do one scenario at a time. The software is specifically suited for utility scale transmission and distribution systems and excels at that. It would not be suited at all for an industrial plant or any type of building electrical system. Small things like impedances within a substation are difficult to model correctly.