Sanderxpander
I second the RME UCX/UFX series by the way. You can get really low round-trip latency with those on a simple USB2 connection. I can't imagine the benefit of Thunderbolt speed-wise there.

If we had proper PCIe via Thunderbolt drivers on the PC:
Like a quality PCIe audio interface, you could go down (effectively) to a 32-sample ASIO buffer size.
That's the only advantage...
 
Early in its life, MOTU allowed the 896HD (and other mkII series) to go down to a 32-sample ASIO buffer size.
I wish they hadn't removed that option...
I'd like to see this available on the better USB audio interfaces.
 

The difference between a 32 and 48 (or even 64) sample buffer, especially with interfaces like RME's which are very fast at the hardware level as well as having very good drivers, is so little it's not really worth bothering about. It's the difference between having a speaker three feet from your ear and five feet.

Thunderbolt really adds nothing to audio processing other than allowing the use of external SSDs with TRIM support, at least at the moment, unless you need a huge number of inputs with live monitoring through software on many of them so would be using PCIe cards otherwise. USB2/3 and Firewire have sufficient bandwidth to handle a lot of channels at current bit depths and sampling rates. This may change of course if "standard" bit depth/sampling rates increase or a new digital audio technology arrives which does things in a different way and needs more bandwidth to do it.

I'm with tlw. The difference between 32 and 48 is so minimal. If that's really significant for you, move closer to the speaker, use a higher sample rate or use direct monitoring.

I get that it is still faster than USB, but most of the time I see people aching for new TB stuff they don't realize how good recent USB devices have become and how small the difference is.

Someone else said it best (can't remember who it was).
"Thunderbolt is a solution to a problem that doesn't exist."
 
RME USB units offer fantastic (low) round-trip latency performance.  
MOTU's new Ultralite AVB is also a great performer (4.9ms at a 64-sample ASIO buffer size/44.1k).
 
I'm all for flexibility.
Give me a 32-sample ASIO buffer size... and I (and the speed of my machine) will decide if/when to use it.
It is getting "nit-picky" when you're talking sub 5ms, but (to me) comparing the feel/immediacy... lower always feels tighter and more immediate.  Especially when playing sounds with strong attack-transients...
 
 
 

It's not just that USB is "fast enough", it's that the TB speed gain is so tiny. We're talking perhaps 2ms here. Which equates to just over half a meter of distance from the speaker. I do get where you're coming from. I just want to make a point against this "I have to have a TB audio interface because it's much faster" craze.

I want earphone distance latency.