MIDI USB Latency/jitter test

[Simon Ravn]

Hi! Out of curiosity I did some tests with an USB MIDI interface and GigaStudio.


Test machine - separate machine for GS only, MIDI connected via Unitor8 SERIAL on a P3 667 MHz:

AMD Tbird 1 GHz, 200 MHz FSB
MSI K7T Pro2-a (KT133 chipset)
1 GB SDRAM 133mhz
MIDISport 8x8 USB
RME Multiface
Harddrives used for .gig\'s: 80GB Seagate Barracuda IV 7200 RPM, 40GB IBM GXP75 7200 RPM, 40GB Maxtor DiamondMax+ 7200 RPM.

TEST 1

24-bit 44.1khz stereo, heavy MIDI orchestration on all channels, instruments loaded on almost all channels in GS. Very heavy orchestraion in bar 1, not quite as
heavy on bar 2.


Track 1 port1 MIDICH 1:

Bar 1, beat 1: 2473 samples = 28 ms
Bar 1, beat 2: 2950 samples = 33 ms.
Bar 2, beat 1: 1569 samples = 18 ms.
Bar 2, beat 2: 767 samples = 9 ms.
Bar 2, beat 3: 1117 samples = 13 ms.


Track 2 port2 MIDICH 5:

Bar 1, beat 1: 2601 samples = 29 ms.
Bar 1, beat 2: 3079 samples = 35 ms.
Bar 2, beat 1: 1825 samples = 21 ms.
Bar 2, beat 2: 767 samples = 9 ms.
Bar 2, beat 3: 1245 samples = 14 ms.


Track 3 port3 MIDICH 13:

Bar 1, beat 1: 2473 samples = 28 ms.
Bar 1, beat 2: 3079 samples = 35 ms.
Bar 2, beat 1: 1825 samples = 21 ms.
Bar 2, beat 2: 767 samples = 9 ms.
Bar 2, beat 3: 1373 samples = 16 ms.


Track 4 port4 MIDICH 12:

Bar 1, beat 1: 2474 samples = 28 ms.
Bar 1, beat 2: 2952 samples = 33 ms.
Bar 2, beat 1: 1569 samples = 18 ms.
Bar 2, beat 2: 767 samples = 9 ms.
Bar 2, beat 3: 1245 samples = 14 ms.


Woah - talk about jitter - from 9ms to 35 ms. 26 ms. of jitter! Fortunately I am mostly doing orchestral music, and I guess you would normally not get this heavy MIDI transmission if doing pop, rock or
whatever. Anyway, I would never record this much MIDI to my machine at the same time. Also because I want to be able to mix all sections of the orchestra
independently - at least divided into strings, brass, woodwinds and percussion - and more likely more divisions than that. So this MIDI jitter will never be as bad
when I go and record to the sequencer machine.


TEST 2

24-bit 44.1khz stereo, medium orchestration on 16 random channels across all 4 ports. A lot of expression, modulation. Some slow chords, some fast chords, a couple of melody lines. Equally dense orchestration in bar 1 and 2.

Track 1 port1 MIDICH 1:

Bar 1, beat 1: 679 samples = 8 ms
Bar 1, beat 2: 645 samples = 7 ms.
Bar 2, beat 1: 799 samples = 9 ms.
Bar 2, beat 2: 765 samples = 9 ms.
Bar 2, beat 3: 603 samples = 7 ms.


Track 2 port2 MIDICH 5:

Bar 1, beat 1: 679 samples = 8 ms.
Bar 1, beat 2: 773 samples = 9 ms.
Bar 2, beat 1: 799 samples = 9 ms.
Bar 2, beat 2: 765 samples = 9 ms.
Bar 2, beat 3: 731 samples = 8 ms.


Track 3 port3 MIDICH 13:

Bar 1, beat 1: 679 samples = 8 ms.
Bar 1, beat 2: 773 samples = 9 ms.
Bar 2, beat 1: 799 samples = 9 ms.
Bar 2, beat 2: 765 samples = 9 ms.
Bar 2, beat 3: 731 samples = 8 ms.


Track 4 port4 MIDICH 12:

Bar 1, beat 1: 679 samples = 8 ms.
Bar 1, beat 2: 773 samples = 9 ms.
Bar 2, beat 1: 799 samples = 9 ms.
Bar 2, beat 2: 765 samples = 9 ms.
Bar 2, beat 3: 731 samples = 8 ms.


As you can see the latency is pretty much the same on all channels on the same beat (at least those 4 randomly picked channels) - only two beats on channel 1
are a bit off compared to the rest. 2 ms. of jitter seems very acceptable to me - of course the figures are not quite as good when the orchestration is a lot denser
and GS is pushed to the 160 voice limit (and beyond). I am not the one to judge whether the MIDI timing is so terrible during heavy MIDI transmissions
because of USB, the host (sequencer) Unitor8 MIDI transmissions from Cubase, the MIDISport on the GS machine, the MIDI driver(s) or the way GS handles MIDI.

At this point I think it\'s very questionable to blame the USB MIDI interface for this. I remember when I ran on a serial and/or ISA MIDI interface, I had delays in a lot of sounds from GigaStudio due to notestealing and probably because the HD\'s couldn\'t catch up.

Maybe I\'ll do another test later with a softsynth or something else installed on the GS machine and see if the MIDI timing is just as bad. I\'d also like to do a test running from Cubase to one of my regular synths and see if jitter happens on the Cubase side of things or not. I DO have a feeling that the HD-streaming could be a contributing factor for cluttering up the PCI bus and getting the samples pulled off at the right time. If anyone has done some similar tests, I\'d be very interested to see them.


Simon Ravn

[Lucas]

Interesting data, Simon.

I believe there are USB ports on some PCI cards and I wonder what a similar test would show using this. Ditto for USB 2.0.

I\'m holding off on USB until more data comes in.

Here\'s an article that may interest some:
http://www.rme-audio.com/english/techinfo/lola/usbkills.htm (\"http://www.rme-audio.com/english/techinfo/lola/usbkills.htm\")

Lucas

[rkmusic]

Simon: Thanks for your valious testing data. I you said similary, I prefer also to make \"takes\" of different sections for mixing purposes. The great thing with GStd is that you can have the whole arrengement and hear it anyway and then grabbing tracks in preferred seq. prog in more than one pass, isn´t it? remember anyway that if you are (your physical ears) at 1m from the speaker you hear the sound with 3ms delay!

Lucas: thanks for the RME link. Yes, nowaday the USB question is no more a problem with modern mobos, I think. Anyway I will have co live with since the new 5.0 Logic version is shipped with a USB \"donglette\"...

Richard Krull

[Haydn]

Simon,

I think the delay you\'re getting is just the serial nature of MIDI. As you send more it slows down. I use to run into flammed chords, tempo slowing down, etc. in dense passages when using my old outboard synths. The timing problems of past have disappeared since I started using GigaStudio with the same sequences. Cakewalk and Giga are on the same machine and don\'t have to go through my MIDI interface. I\'ve wondered if I\'ll start having timing problems with a separate GIGA machine. The 20 year old MIDI interface spec needs to go!! I\'ve been waiting a good 10 years for its replacement.

[Cool7s_Dad]

Haydn,

I think you\'re spot on about this. MIDI communication is only 32kbps. That\'s about 900 note-on/note-off messages per second. Throw in some controller data and there\'s no question that it\'s too slow. I know there are some less than wonderful utilities like MIDI-Via-Net out there, but what\'s really needed is some sort of straight USB connection from one computer to another. USB would have the capacity to support tens of thousands of messages per second. Perhaps one of us needs to work that problem...

Peace,
Tim
http://www.elithic.com (\"http://www.elithic.com\")

[Haydn]

Actually there are a few companies with some possible solutions that cover transfer of all sorts of data through one cable. This includes digital audio, MIDI, network data, etc. Just need the companies to come up with a standard.

I was just looking at Gibson\'s MaGIC which uses standard ethernet cables. This technology can be used in everything from home automation, live performance, studios, etc. Check their website for more info. It looks quite promising and will be able to use current structured wiring in houses without having distance problems of other systems because it uses standard ethernet cabling.

[ellle]

audio networks
http://www.prosoundweb.com/industrybiz/commentary/chrisd/aes.shtml (\"http://www.prosoundweb.com/industrybiz/commentary/chrisd/aes.shtml\")

[Haydn]

Here\'s the link to Gibson\'s MaGIC system: http://magic.gibson.com/index.html (\"http://magic.gibson.com/index.html\")