I have a large session which involved many stems being imported into the Sonar session. (over 60 tracks but lots of silence)
 
This may sound like a really dumb question. But if I edit every tracks clips to just where there is actually audio and remove the sections with silence with this decrease the CPU usage(in the sections with no clips) if I have effects in the FX bin?
 
It seems to me like it should as there would be nothing for the FX bin to process with no wave file there at all.
 
Did I just answer my own question?

In general, DAW should call plug-ins even in case there is no real waves, just with zero signal, till the plug-in is disabled. Even when the transport is stopped.
There are exceptions, but that depends from concrete FX plug-ins.
PS. Sonar does not have detailed performance meters, but if you open the project in Reaper (there is my extension which allows that), you can check which plug-ins/FXbins consume too much and use that information to temporary freeze corresponding tracks in Sonar.

And VST3's *should* go into reduced CPU usage mode when they are being fed silence (as should some older DirectX ones!).

KPerry
And VST3's *should* go into reduced CPU usage mode when they are being fed silence (as should some older DirectX ones!).

VST3 specification is less strict then you
VST3 host CAN provide silent flag and VST3 plug-in CAN use that flag when set. In practice, good written plug-ins are detecting silence even in VST2 incarnation and such detection is tiny compare to the rest of operations required.  Bad written plug-ins are free to ignore the flag. So from all reports I have seen, even in case the host and plug-in take the advantage of this flag, there is just subtle effect on total performance.
 
But that is a good point from other perspective: if "silence" in the original material is not "digitally dead silence" but just background noise, deleting it will instruct Sonar and plug-ins there is nothing to process. And that can (still plug-in dependent) greatly impact the performance.
 
Note that some "vintage" emulators can generate some "very special vintage noise" even when they have dead silence on input. All future FXes in the chain are forced to process that noise as usual.

Unfortunately, the code within most plugins continues to run during silent periods, even if it's just processing zeroes. It just doesn't have any way of knowing whether the "silence" is truly silence, and strictly speaking it often isn't. The only way to guarantee zero CPU usage is to render those tracks.
 
The good news is that freezing and un-freezing tracks is quick and painless. Even if you only freeze those tracks that you're pretty sure won't need additional tweaks, that can make the difference between your CPU being able to keep up or not. If you later realize you do need to make effects changes, you can get the original track back in the blink of an eye.
 
At the risk of stating the obvious, you only need to take CPU-conserving actions if you're getting dropouts. Your system can either keep up or it can't - there is no benefit to reducing CPU usage if data is already making it out to your interface fast enough, even if the meter shows a frighteningly-high amount of CPU usage. And this is assuming you've already set your buffers to their maximum, which you should be doing as a matter of course. Latency is a non-issue when mixing stems.

Thanks for the responses ! I will freeze more tracks! 

With your rig you shouldn't need to freeze anything