USB has incredibly high latency, and more importantly, non-deterministic latency
Seeing that RME's USB 2 interfaces manage to stream like 50 or more 24bit audio channels at 48kHz with buffer size small enough to get latencies in the mSec ranges, I always wonder: are other manufacturers just doing it wrong? I know that doesn't completely cover the non-determinism argument, but 'incredibly high' seems to be covered pretty well.
The RME-Audio devices do indeed have low latency, at the limits of the USB 2.0 spec, 125 microseconds I believe. They crank up that USB poll rate. And are also using the Arasan chipset IIRC, the same to be found in some of the other prosumer and pro line-up of equipment, e.g. the Solid State Logic h/w. I am hazy on the details, it has been a few years since I was inside any of those devices, people from RME and SSL please feel free to correct me as to your chipsets. Some are using dedicated FPGAs to handle the data capture and processing before handing off to USB. RME's devices are definitely doing a bunch of on-board processing before giving it to the USB bus, and making sure the packets going out are as small as can be. Most non-integrated USB controllers are using VIA or Renesas. USB 2.0 has lower latency but less consistency (shared bus) vs USB 3.0 which has higher latency but is more consistent (point-to-point protocol). Obviously you don't want to go sharing your USB 2.0 port on your PC with an RME and a bunch of USB 3.0 devices, e.g. an external drive, because then you just end up with the worst of both specs, terrible latency and terrible consistency.
Seeing that RME's USB 2 interfaces manage to stream like 50 or more 24bit audio channels at 48kHz with buffer size small enough to get latencies in the mSec ranges, I always wonder: are other manufacturers just doing it wrong? I know that doesn't completely cover the non-determinism argument, but 'incredibly high' seems to be covered pretty well.