Efficient routing for iterative processes

So until recently my idea of efficient coding in Reaktor has been the fewer modules needed to achieve the end result the better, but lately I have questioned this approach upon noticing heavier CPU strain with increased voice count. For example, if I were building an additive synth with partials at 1x, 2x, 3x... the fundamental frequency, my first instinct would be to simply set the voice count to correspond with the desired # of partials and multiply the frequency with the V output for the corresponding voice, though apparently it's actually more CPU efficient in practice to do all these calculations one by one with separate modules. Is there some way to do achieve this process with the iterator module using a single voice that I'm overlooking? I already know I could just use the Sine Bank oscillator for this kind of thing, but I'd like to retain finer control of the individual partial shape, and I'm interested in how to approach the concept going forward outside of that particular use. I'd really prefer not to be manually feeding values to a bunch of modules if I can help it...

Find more posts tagged with

Comments

colB

There is no one best way to do this kind of thing. It depends on the details. Sometimes its more efficient to use polyphony hacks, sometimes it's better to just have lots of oscillator instances in a core cell… (Sometimes there are even tasty math tricks that mean loads of partials can be had basically for free….)

The only way to know for sure is to build using different approaches and test the cpu usage.

It's rarely good to start with optimisations… better to build the thing, then start searching for ways to streamline the cpu load once it's making the right noises.

silvermoats

https://community.native-instruments.com/discussion/comment/235122#Comment_235122

Fair enough- I guess it comes down to whether we're defining effiency in terms of processing or ease of assembly - I'll try to work my way backwards from the latter as a starting point.

Were you alluding to multiplying sines at different offsets exponentially in Core or something else in particular you'd care to elaborate on?

colB

https://community.native-instruments.com/discussion/comment/235124#Comment_235124

Frequency modulation, Amplitude modulation, phase modulation, basic stuff like that gives extra partials. Chowning style FM really does give a lot for very little. chebyshev polynomials are interesting… Moorers discrete summation formulas… There is a lot of interesting stuff out there, most of it from decades ago :)

silvermoats

https://community.native-instruments.com/discussion/comment/235129#Comment_235129

Gotcha - I've experimented with most of those but have only been able to get independent control of individual partials through AM-based processing. Will dig more into the other methods as well, thanks for the ideas

gentleclockdivider

yes , use the iterator
Preferable combined with an event able , this way you can morph between y-locations and it's great fun , especially with the modal bank .

In a nuthsel , the first ouptut of the iterator should go to the index value of sine -modal , then and only then you should adres sthe other inputs .

You confirm the changes with append

https://i.imgur.com/ORXX5lR.gif

https://i.imgur.com/EWixXjW.gif

Quick links

Unanswered questions