Pierre-Olivier Boulant

WFS_24I_46O_6R.PNG

WFS_24I_46O_6R_avx.PNG

This is a test of the audio engine using a poly~ and keeping one voice (thread) only for IO and the rest for the processing. I'm pushing it a bit with the channel count I have 24 inputs and 46 outputs plus 6 reverb sends and returns. This is running on a i9 9900k@5GHz on all eight cores with plenty of cooling and a RME MADI-fx @96kHz.

If you look at the pictures all I've changed is the DSP vector from 128samples to 64 and the CPU% drops by half while the CPU speed goes grom 5GHz to 4GHz. This is the motherboard dropping the CPU frequency because AVX2 kicks in and this is usually produces more heat on the chip.

In the end my question is this: how can we consistently enable AVX and not stumble upon it at a certain setting. Because I tried other settings (input, output count and IO vector, DSP vector) and you can never be too sure why and when it is enabled or not.

Spot the difference!
 This is a test of the audio engine using a  poly~ and keeping one voice (thread) only for IO and the rest for the  processing. I'm pushing it a bit with the channel count I have 24 inputs  and 46 outputs plus 6 reverb sends and returns. This is running on a i9  9900k@5GHz on all eight cores with plenty of cooling and a RME MADI-fx  @96kHz.
 IO vector is at 256 samples.
  If you look at the pictures all I've changed is the DSP vector from  128samples to 64 and the CPU% drops by half while the CPU speed goes  grom 5GHz to 4GHz. This is the motherboard dropping the CPU frequency  because AVX2 kicks in and this is usually produces more heat on the  chip.

In the end my  question is this: how can we consistently enable AVX and not stumble  upon it at a certain setting. Because I tried other settings (input,  output count and IO vector, DSP vector) and you can never be too sure why  and when it is enabled or not.

 
 

 
Spot the difference!
 This is a test of the audio engine using a  poly~ and keeping one voice (thread) only for IO and the rest for t