1 Topic by r.kemp

  • r.kemp
  • Member
  • Offline
  • From: Birmingham, UK
  • Registered: 2008-11-25
  • Posts: 17

Topic: Redistribution of DSP load

Hey,

I'm working on a system profile that has 5x N6 and has 2x CAB (16o & 8o). The Processors are split up over two AV Rooms. 3xN6 on the south side and 2N6 & CABS in the North. The bulk of the audio input sources are located in the south side across the 3 N6's. I've just tried to add additional processing but it appears that the 1st DSP is doing all of the work while the others are relaxing.

I wanted to know if anyone can recommend a way of distributing the DSP load across all of the DSP better or if anyone has any ideas in general to help ease the mental strain I'm now having :-/ Would locating all of the DSP in one rack and then using CAB for Audio break in and out at the relevant North /South rooms be a better option?

Thanks for even stopping to read this, any input is appreciated smile

Richard Kemp

It already sounds better

r.kemp's Website

2 Reply by JoshM

  • JoshM
  • JoshM
  • Recording Engineer at Burnt Orange Studio
  • Offline
  • From: Meridian, MS
  • Registered: 2007-08-13
  • Posts: 1,498

Re: Redistribution of DSP load

Hi Richard,

The best way for us to offer suggestions would be to have a look at your project file.

Unfortunately, we can not attach files here (something we are looking to change), but if you would like to email it to me I can look it over and get back to you.

Generally speaking (and this is very general) the compiler attempts to use all the resources available on one DSP before moving on to the next DSP. So, if you have a NION n6 and you are doing very little processing with it, all of the processing may be done on just the first DSP (it could be at 80-90%) while the rest of the DSP's are not doing anything. This is totally normal and results in the lowest latency through the system. After all, the fewer times the signal has to be passed between DSP chips, the faster it gets through the system. However, this talk about latency through the DSP should be qualified by the comment that the latency through the DSP process is by FAR swamped by the latency through the ADC on the front and the DAC on the output.

Here is a fun thing to try:
1. Set up a NION with an analog input and an analog output
2. Configure the DSP to take the input directly to the output without doing any processing to it.
3. Using something like Smaart or your measurement tool of choice, measure the delay propogation through the box and write it down.
4. Now add a whole bunch of processing devices between that same analog input and analog output.
5. Measure the delay propogation through the box again and compare it to the previous value.

When I did this with Smaart 5 there was very very little change to the delay propogation through the box. In fact, when I measured it, Smaart was not precise enough to find any difference in the delay propogation and I think I spread the processing amongst the DSPs in the box so it would have to jump from chip to chip.

Something I would like to try sometime when I have a little time is to do the same thing as above with both the analog I/O and the NIO-AES card to provide digital I/O. However, at this time, I do not have AES or SPDIF I/O for my computer. This would be a good way to illustrate the differences between the analog and digital I/O schemes. Alas, there is just not enough hours in any given day to get all of this done.

Josh Millward
Burnt Orange Studios

JoshM's Website