The simplest thing to do of course is just try it. You can install Windows in a VM without a product key and try it without limitation for something like 30 days.

That said, "really low latency" is relative. Device assignment will give you excellent I/O performance in terms of sending data to/from the device, particularly for bulk DMA transfers. Where you (likely) still pay a penalty is in terms of interrupt delivery, as on most chips these are still routed through the host even in "passthrough" mode. Even so, you're looking at delays at the outside on the tens of microseconds scale, depending on which microarchitecture you're on and how well tuned the fiddly bits of your hypervisor (presumably KVM for most users) are. Some configurations will be much better behaved here. If your audio hardware goes into polling mode (or can be made to do so) when it is recording you can potentially avoid the interrupts entirely.

(source: I work on Google Compute Engine, in particular on virtualized networking, historically on performance. Virtual interrupt delivery latency is a topic near and dear to my heart, although I don't know a damned thing about audio gear and its tolerances there :)

I personally install KXStudio[0] for guys just like you almost monthly. here's some appropriate[1] pornography[2]. Imagine being able to have a skype call with someone, and be able to record their vocals (provided they have a good mic) apply effects, and then record _and_ pipe the vocals back to the vocalist in realtime. Or make recording snippets of audio off youtube trivial. Or stream live instruments in other rooms over wifi/ethernet using only a raspberry pi and a good microphone. Or stream 8 of those tracks/instruments (on different devices) to one computer for mastering/final mixing. It's audio nerd nirvana, basically.

What you need: Wine Wineasio (for Wine support in FL Studio or others) KXstudio

1. http://kxstudio.linuxaudio.org/index.php 2. https://www.youtube.com/watch?v=GBD-wsMiZCk 3. https://www.youtube.com/watch?v=PhfmGGPcwNc

For lack of a response from someone who has actually tried getting ASIO working on such a setup, I suggest you just try it. Grab a Live CD and a spare external hard drive and just give it a whack. Passing through the sound cards should work. USB might be trickier if you're not just passing through a whole USB hub. But considering how new IOMMU is, I imagine the intersection between people who have tried it and people who use Cubase is incredibly tiny.

I have found that a lot of the audio editing software you can get away with quite low specs. I record my podcasts on a 8 year old laptop running Windows XP. It rarely connects to the internet and I record to a thumb drive that is backed up so there is nothing to lose on it. In this particular case something similar might be the best option. For pod-casting anything over 2 or 3 ms in my headphones throws me off my game.

Unlike podcasts, music recording and mixing is extremely CPU intensive. Especially if you run a lot of analog simulating synths and stuff.

Virtual CPUs with VTx can be made to run at essentially native speed for most workloads on most fairly modern CPUs. Memory performance can be subpar, although there's public documentation on techniques that can improve that: https://help.ubuntu.com/community/KVM%20-%20Using%20Hugepage...

I/O performance is a whole other story, dependent on many workload and hypervisor specific factors.

I stand corrected.

Really, the 965 supports it? I have the 970 and was under the impression the Phenom II's in general just don't support it. Can you show me where you found that information?