Octo Capture, RME Fireface UC
If lowest possible round-trip latency is important to you:
RME USB units yield 4.9ms at a 48-sample ASIO buffer size/44.1k
OctoCapture yields 7.4ms at a 48-sample ASIO buffer size/44.1k
Presonus 1818 VSL yields 4.9ms at a 64-sample ASIO buffer size
MOTU hybrid units are ~5.5ms at a 64-sample ASIO buffer size
The above at the best performers when it comes to lowest possible round-trip latency in a USB unit.