I've took some code from someone else which name I don't remember (sorry man!) but I only use this script for private use. Still I need to share it here for someone to look at it.
It's a speaker that reacts to bass and gets bigger and shakes. I've got two of them, the other one is for high frequency sounds. As soon as I start one up my CPU usage goes up by 30%, using both doesn't make much difference. I tried to set down the fft size but it didn't do much. Here is the code:
EDIT: I forgot to add that I am using 2 spectrum analyzers aswell but they don't use much CPU, maybe 5%. My CPU also isn't very bad, it's mid-end (Intel i5)
Sound and audio visualization skins are notorious for being CPU hogs. On my i7-3630QM this skin 7-10% of my CPU which seems about right depending on which i5 you're using. The best way I know to get better performance in visualization skins is to increase Update= under [Rainmeter].
Update=10 is what you used and eats 7-10% of my CPU. This setting also causes my i7 to start increasing it's MHz speed above the minimum 1197MHz. Update=50 still looks pretty good but eats 5% of my CPU. This is probably what I'd recommend trying. Update=100 looks a bit choppy but eats only 2.5% of my CPU
At the end of the day you'll have to decide what balance of performance and quality you like best.
I'd just add to what exper1mental wrote, that in fact the Update=10 is a totally wrong value. The lowest possible value for Update is 16 (read the second sentence here). With Update=10 the skin will work the same way as with Update=16.
If your sound card runs at a very high sample rate (96khz, 192khz) that will also bump things up quite a bit. Also if you're running your card in 7.1 vs 2-channel audio.
If all you want is a speaker throbbing, you might try using RMS levels rather than doing a full FFT on the audio stream.
Just read your edit - are you saying you're running 3 separate parent AudioLevel measures? Keep in mind that each one of those is requesting a full stream of all the audio data from the sound card, which is a pretty expensive thing. You generally only ever want one Parent AudioLevel measure per machine.
Also in the skin you posted, you're doing a 1024-point FFT, and then interpolating it into 3 bands, which basically does a lot of math and then throws away the results. If you only need 3 bands, try a smaller FFT. (or change it to RMS which is a lot cheaper)
