[eon1]

> low latency and aggregated devices

Hang on, no way, you take one or the other. Aggregating devices adds a huge chunk of latency and, frankly, I don't think is that exciting a feature anyway. Maybe for your specific setup, but generally speaking you should buy gear that suits your needs, rather than try to cobble something together from existing devices.

Having used OSX and Win7 (on the same machine) I would agree CoreAudio is less hassle and definitely lower latency - IIRC, OSX reported just over half the latency of W7 on the same setup.

That said, I'm not sure what you think is going on with ASIO, since any class compliant USB audio interface has compatible drivers inherently..

[eropple]

> Aggregating devices adds a huge chunk of latency

In the situations where I have to do it--and it's less about "buying the gear that suits my needs," more "having to bash together other people's gear on a shoestring"--I find that aggregation ends up around 30ms, which is on the higher end of acceptable when monitoring in-ear. (To be specific--this is not for music but rather audio routing for video. Occasionally I'm on-site somewhere and need to be able to monkey up something a little faster than I'd like, you know?)

Trying to do that with ASIO at all is impossible. So it's a pretty big deal.

> That said, I'm not sure what you think is going on with ASIO, since any class compliant USB audio interface has compatible drivers inherently..

In theory, yes. In practice, I find ASIO a little unstable (my home setup uses Ableton on Windows as a live mixer through a TASCAM US-16x08, though I'll be going back to using a Mac when I get back my Mac mini from a friend). I have never had Core Audio kernel panic a machine, but I've had machines (in one memorable case, the same machine dual-booting a hackintoshed OS X and Windows) hard-lock as soon as I enabled ASIO on two different devices (my old TASCAM US-1200 and my a friend's 8i6).

It is not unusable, by any means, but for my purposes (again, live shows) predictability is a big plus.

[eon1]

> I find that aggregation ends up around 30ms, which is on the higher end of acceptable when monitoring in-ear

I haven't done that kind of work, so for video/speech, fair enough. 30ms (on top of whatever's already there) is not really acceptable for a musical performance, though - the performer will be expending half their brainpower trying to mentally align what they hear through bone conduction vs in ear monitoring and adjusting their performance based on outdated/scrambled information. It's a really confusing experience - if you've ever seen that Japanese "speech jammer" device[0] you'll get some idea what I mean (although it uses much longer than ~30ms, that's still enough to mess with you).

Working in a pinch, it would make a useful and more stable tool than ASIO, though I run either analog or dedicated devices for any live work - I just don't trust computers that much - but I get that working with video means you might not have those kind of options. I can't say I've had many of the troubles I've heard about with ASIO to be honest, but I undeniably do enjoy working on OSX more anyway. This conversation reminds me how badly I need to revive my ML Hackintosh..

[0] https://www.youtube.com/watch?v=USDI3wnTZZg