Interesting CPU saving discovery

[ThosSounds]

Hi all,

I just discovered something I'd not heard about before in Pro Tools HDX and thought it might be useful to some (and maybe someone wiser than I could even explain it!)

Using Pro Tools HDX, I've been running some hi-res sessions at 32bit, 96k. It's fairly easy to push my old 2010 Mac Pro to its limit on this. On several sessions I've been pressing the CPU hard enough with native-only plugins that I wind up having to freeze tracks, etc to save CPU. My master fader is often filled with 4 native plugins: FF Pro-Q2, Waves S1, FF-Pro MB and FF Pro-L.

Today I discovered that if I insert a DSP plugin in-between two of those 4 natives, my total CPU use bar drops by about 25-30%. Putting the DSP plugin before or after the natives has no such effect, it has to be in-between them (something I thought it was frowned upon to do?) Delay compensation shows the master fader in orange amounts either way but it shows a much higher delay with the DSP plugin inserted. Any DSP plugin seems to work, it can even be bypassed (though not "inactive")

Bizarre trick that seems to be working for me. Hope it helps someone else!

My specs just FYI:

Pro Tools HDX 12.4.0 on Mac OS 10.11.6, Mac Pro 2010, 2x2.4GHZ Quad Core Intel Xeon, 24GB 1066 MHZ DDR3 RAM
All plugins are on their most current versions (as of today's date :)

Thanks and good luck!
Tom

[Drew Mazurek]

A wild guess......

since it goes between the native plugs (leaving the DSP world and coming back), which puts them into different "processing streams" so to speak, they get handled on a different core (or virtual core?), whereas when all the native processes happen in serial, they are in the same stream and are concentrated/ confined to the same core?

I have no idea what I am talking about, does it show???!!?

[panamajack]

Originally Posted by Drew Mazurek

. . . they get handled on a different core (or virtual core?), whereas when all the native processes happen in serial, they are in the same stream and are concentrated/ confined to the same core? . . .

Makes sense to me...

[Darryl Ramm]

And the latency/delay increase should just be due to another round trip through the IO buffer plus any DSP plugin latency. I would just not worry about what the meters show, especially with hyper-threading they demonstratively show nonsensical behavior.


Sent from my iPhone using Tapatalk

[LukeHoward]

I accidentally discovered the same thing recently, although I used cascading aux buses (which, if you think of the mixer as just another DSP plug-in, is a special case of what you describe). It does seem this affects how the cores are allocated to native plugins.