It's alright. Right now 2.5ms is not bad, but in 10 years it might be part of the conversation when choosing consoles. A 5D has 2.31ms and the AVID MixRack from Avid has <2.3ms. The only company killing the
latency issue is DiGiCo at under 1ms. A friend of mine here at school is working in a lab that has made a AD/DA that can operate at 35 microseconds(0.035ms). Now it can't convert anything close to real sound, but the work is in the pipeline for major improvements. If
Yamaha is going to have this as their flagship for too long that number might become a problem, but not for a while.
The issue is not just A/D and D/A
latency but also
latency associated with the processing and routing. In fact the A/D and D/A
latency may by a small portion of the overall
latency.
You also have to consider if the consoles incorporate some form of delay compensation. A simple example, split a mic to two channels. The A/D
latency should be the same for both channels but if one
channel is heavily processed and the other is not then the processed
channel will have additional
latency associated with the processing. If you then combine those two channels on a bus without delay compensation you would have the same signal with
phase differential being summed. So, do you allow that? Or do you calculate an absolute worst case potential
latency and then apply
latency as required to each
channel so that every
channel has that same
latency? Or do you constantly calculate the
latency for every
channel, determine the worst case and apply an appropriate compensation delay to each
channel so that the
latency on every
channel equals the worst case at any time? A similar situation applies wherever channels or buses may sum and to the outputs (should
latency from any input to any output be the same regardless of the path or processing?).
This can get into tradeoffs and each manufacturer will use the approach that works best for their
console architecture. Assuming a potential worst case
latency may mean assuming a longer
latency, but can simplify the math. Keeping
track 'real time' of the delays and applying constantly varying delays may allow accommodating shorter 'worst case' delays but requires more processing to do all the related math in 'real time'. Not applying delay compensation or maybe just an overall input-to-output compensation is less complicated and should result in lower
latency for most paths, but could be problematic with concepts such as parallel compression where you split a source to two inputs (or an input to two buses), compress them differently and then sum them back together.