View Single Post
Old 19th September 2019
  #43
Lives for gear
 

Quote:
Originally Posted by psycho_monkey View Post
Even a proper trip through the software at 32 samples is “worse” than dsp use. And you’d need a computer powerful enough to run that buffer setting at all times - because right now you can have a mix straining with native plugs and still punch in at low latency over the top.
From what I've gathered from @ ProPower who's measured with a function generator-

@44.1/48 HDX is 1.9 with only an audio track, 2.5ms with one send and submix and ~10-100 samples per plugin monitored through, so depending on plugins 2.7ms to 4.7ms. HDN is 3.35ms at the 32 sample buffer and most plugins add 0 samples.

@ 96k HDX is .45 only an audio track, ~1ms with one send and bus and plugins adding from .1ms to another 1ms, so 1.1ms to 2ms. HDN is 1.8ms @64 buffer.

But HDX also increases RTL with more auxes and busses...not by a lot, but in these margins even an .5 ms increase is significant. Adding submixes and auxes has no effect on RTL for native. Maybe ProPower will chime in with more details.

Of course it's true that you have to run the lowest buffer settings, but this is no issue on my 7980xe systems if the sessions don't have VI's or if you freeze the VI's. I wish there was an option to have double or even quad buffer settings for VI's in PT.

It is true that CPU use goes up dramatically when you go from the 128 buffer to the 64, from 64 to 32 isn't as dramatic, so there is some threshold there (at 44.1/48). From what I can see tinkering with 2caudio breeze, support for AVX 512 seems to make a solid improvement there...IE less of a jump going from 128 to 64 and then 32. I wasn't able to test it before and after though so this is a bit conjecture. I will say that at the 32 sample buffer plugins have to be coded well, some plugins, in spite of having plenty of CPU overhead, will not run cleanly at the lowest buffers.