DPC latency is the max delay ( (usually from poorly written kernel level driver interrupt code) ) resulting from displacing
what you might want executed at a lower priority (like audio transfers, although those are also part of the same priority queue) into deferred calls
(as opposed to things like graphics painting or memory mgmt)... With a priority queue for DPCs where all of these get placed for processing (not sure how concurrency/sync/starvation are managed there, but there are definitely things you want favour ASAP execution of vs audio
Hence DPC is not really absolute system latency.
Rating $k/ms latency is just ridiculous. It's about as of a legitimate or useful heuristic as saying an expensive food joint is necessarily good. Just because there's a
high margin, doesn't mean people spent any time designing the DAC section or minimizing magical latency (presumably he talks about inherent jitter of the various interface components (dodgy CD opt out to dodgy dac chip opt in ?).
Yes, totally. And people often confuse software latency with hardware. Software latency issues means delay between a program calling an API function and the output of the drivers and has nothing to do with the internal electronics.