Its a combination of USB, host, shitty drivers on sub-par hardware. I did some studies when I had a number of different interfaces to test with a Core8 as a measurement device and here are the results:
(TS= Timestamp on, DS= Directsound)
The RAW is a physical connection from in to out (which should be the length of the midi msg), the midiox test is a basic software loopback
You can see how much everything changes with the host as well as the driver. Most interesting though is how the Midex (which was a $450 interface at one time) degrades with more data and how the whole “LTB” (timestamping) is utter bullshit
Thanks for shedding some light on the topic! (…and very thorough measurement)
The GM5 seems to do a good job, special. not spreading so much. I am just wondering if my (not finished yet) SeqLiteV4 can act as a Midi interface with the same performance… Sonar is crap in that regard then, I am using FLSTUD and that needs to be checked
For future measurem, how do you get to microseconds? what’s the math behind that?