CPU Usage - When and when not?

"Vettetech"
Do soft-synths and associated tracks that have been bounced to audio and then muted still use CPU?

Yes.

"Vettetech"
Does archiving them or freezing them reduce any CPU Usage?

Yes, but not completely. You're basically transferring the load onto your hard drive. And there's a limit to their streaming ability.

"Vettetech"
If an Effect is assigned to a track or Bus but is bypassed does it still use CPU or only when enabled and processing?

Only when enabled and processing, and aside the minimal system resources it reserves for itself for being there.

"Vettetech"
Do Midi\Softsynth tracks use more or less CPU then the bounced audio track?

Did you load the corresponding sample library into memory or are you streaming from disk? If the former, the CPU takes the load (and that's a good thing) and if the latter, it's your hard drive that does (less efficient than the former).

"Vettetech"
is it best to wait to bounce them until ready for the final mixdown to reduce the amount of CPU being utilized?

I would only bounce freeze if I had no choice and was done with the processing for that track. Freezing/unfreezing tracks to make adjustments is time consuming. And you will tax your hard drive very quickly by bouncng/freezing.

IMHO, it's a balancing act as I've learned by working with EastWest's sample libraries. My goal is to load the memory as much as possible without letting CPU usage go above 50%. Then, it's up to the hard drives to deliver. Usually, distinct and seperate hard drives for the audio playback (frozen/bounced tracks) and streaming samples.

For example, EastWest's Bosendorfer 290 with 2 mic positions (approx. 12k samples) is streamed from disk (I have no choice as 16GBs of memory would only take care of approx. 20% of the samples), at a reasonable disk load of approx. 2%. Add to that vocals from VOP, and a couple of cellos which are all loaded into memory, also at a reasonable memory cost of 4% of my 16GBs. But the CPU is taking a bigger hit. Bottom line is we're all trying to avoid the crackles and pops and dropouts and... etc., within the boundaries of what our systems can deliver.

Hope this was helpful,
jc

In the case of loops, my first question would be, are the loops midi based or waves? If midi based, then until I decide to either freeze or bounce the track, it's all memory work. I actually prefer not to freeze/bounce my Superior Drummer tracks, as reading approx. 10 tracks from disk is much more demanding than leaving them loaded into memory (in my case, it can easily add an additional 15% of disk load, which is a lot for 1 virtual instrument).

Try it, do a test. You can open the "Task Manager" as you're doing this. Start a new project with drum loops only (as you normally do) and check the numbers under different scenarios. Should be very interesting as to how your system behaves under different loads.

Let me hopefully add to some of the good information already here...

First, almost all soft-synths are going to have built in effects, especially reverbs. Reverb can be CPU intensive and it is something you may wish to turn off inside the soft-synth. Also, if you have a bunch of different soft-synths with different notions of a reverb space you are in then it may be harder to get everything to hang together regardless of your CPU situation.

One other thing along the lines of effects which is related to above is to take a look at your per track effects, again concentrating on reverb as my example. Make sure you don't put a reverb on each track unless necessary. Instead, create a separate bus for reverb, put the effect on that bus and then route tracks you want reverb on as sends to the reverb bus. WAY more efficient use of your CPU.

I would classify synths in three categories. The first uses the CPU alone to generate sound. Not much of a strain on disk and generally light on memory also. For simple patches like an lfo modulated filter sweep of a square wave you are not going to have problems just leaving the track active and not frozen. Second is a hybrid model of samples plus algorithms. Omnisphere is a perfect example of this. You will use a lot of disk bandwidth if you have large samples and can get crazy CPU useage if you have sophisticated custom modulations. The third type of soft-synth is something like Ivory 2 pianos which has very large samples with lots of depth for realistic piano. Not much of a CPU strain but potentially a memory and disk bandwidth hog.

In many cases you can relieve a lot of strain on your system by doing sample thinning for case 2 and 3 above. Example in Omnisphere is you can dramatically reduce memory use (and system overhead) by giving a range of keys you are going to play for a given patch. Why load up 88 keys worth of sample space if you are playing a single octave of it for example. Ivory allows you to use lighter versions of samples and, in any case, won't load samples unless you play in the range where the samples are. Or so they say...;-) Kontakt libraries definitely would be a candidate for sample thinning. I am not a Kontakt wizard though. Usually, the documenation on the libraries will give you clues to what features of a library are going to be resource intensive.


Just some ideas to consider that may help...

Ken