1 Topic by mxa (edited by mxa 2012-11-30 20:23:27)

  • mxa
  • ADAT user
  • Offline
  • Registered: 2012-11-25
  • Posts: 20

Topic: Latency tests with the Fireface UCX and Firewire vs. USB vs. CC Mode

I did some latency tests with my brand new Fireface UCX. In the software (Pure Data) a buffer of 20ms is defined, so that needs to be subtracted from the  numbers below. The latency is measured round-trip: a cable is connecting out to in (one channel). The bandwith isn't considered in the test, so the Firewire might still be a good choice for multichannel, high sampling rate use cases, but for low latency the USB port is king. Or go for Linux, where the overall system overload seems to be much lower than on OS X. Windows latency sucked.

Tested with Pure Data 0.43.4-extended-20121101
/doc/7.stuff/tools/latency.pd
Software settings: Sampling rate 44100 Hz, delay 20 ms Blocksize 64, No callbacks

OS X 10.8.2  Mac Book Pro 2 Ghz i7
Latency of internal laptop soundcard (Intel High Definition Audio) for reference:
24.348
24.3098
24.3123
24.3148
24.3173
24.3199
24.3224
24.3249
24.3275
24.33
24.3326

Latency Fireface UCX via Firewire, RME Driver
27.8751
27.8777
27.8803
27.8829
27.8854
27.8472
27.8497
27.8522
27.8548
27.8573
27.8598
27.8624
27.8649
27.8675
27.87
27.8726

Latency of Fireface UCX via USB, RME Driver
23.0024
23.005
23.0076
23.0101
22.9719
22.9744
22.9769
22.9795
22.982
22.9845
22.9871
22.9896
22.9922
22.9947
22.9973
22.9998
23.0024
23.005

Latency Fireface UCX via USB, Class Compliant Mode
24.3173
24.3199
24.3224
24.3249
24.3275
24.33
24.3326
24.3352
24.3377
24.3403
24.3429
24.3454
24.348
24.3098
24.3123

For comparison:

Ubuntu 12.04 Intel® Core™ i3 CPU 550 @ 3.20GHz × 4. System not set up / optimized for audio, no special real-time kernel.
Latency HDA Intel (hardware) – the on board soundcard
21.9012
21.9037
21.9063
21.9088
21.9114
21.914
21.9165
21.9191
21.9217
21.8834
21.886
21.8885
21.891
21.8936
21.8961
21.8986
21.9012
21.9037

Windows 7 professional, same Hardware as above
Latency Fireface UCX via USB, RME driver
(USB Class Compliant mode did not work Plug and Play on Windows)
(had to set the delay in Pd to 120 to avoid drop outs. Subtracted 100 from the results below for comparison, were 233.xxx originally!!)
133.917
133.879
133.881
133.884
133.886
133.889
133.891
133.894
133.897
133.899
133.902
133.904
133.907
133.909
133.912
133.914
133.917
133.879
133.881
133.884

2 Reply by MC

  • MC
  • MC
  • Administrator
  • Offline
  • From: Germany/Thailand
  • Registered: 2007-04-06
  • Posts: 24,823

Re: Latency tests with the Fireface UCX and Firewire vs. USB vs. CC Mode

Your first paragraph is already going into the wrong direction. USB shows the lowest numbers because it has the smallest safety buffer, and works in a special RME mode that even outperforms CC (which is as direct as it usually could be). Nevertheless, completely independent from  bandwidth and multichannel use, FireWire will work much better (means allow lower latency settings=smaller buffer sizes) on some machines than USB. And even if FW and USB work perfectly, it might be that USB needs to be raised one step in buffer size to reach the same click-free operation as FW in real-world operation. Or the other way round.

So basically the numbers you listed do not say anything useful, sorry.

Regards
Matthias Carstens
RME

MC's Website

3 Reply by mxa (edited by mxa 2012-11-28 11:00:17)

  • mxa
  • ADAT user
  • Offline
  • Registered: 2012-11-25
  • Posts: 20

Re: Latency tests with the Fireface UCX and Firewire vs. USB vs. CC Mode

The only thing I've measured above is the latency added by the system setup: Operating System, interface and so forth. I think that's already a valuable information, because the safety buffer always comes on top of that, but it is impossible to go below that. How would you design a useful test?

  • vinark
  • DAW Mastermind
  • Offline
  • Registered: 2009-06-05
  • Posts: 2,647

Re: Latency tests with the Fireface UCX and Firewire vs. USB vs. CC Mode

http://www.dawbench.com/ or http://forum.dawbench.com/forum.php As useful as it gets...

Vincent, Amsterdam
https://soundcloud.com/thesecretworld
Babyface pro fs, HDSP9652+ADI-8AE, HDSP9632
  • mxa
  • ADAT user
  • Offline
  • Registered: 2012-11-25
  • Posts: 20

Re: Latency tests with the Fireface UCX and Firewire vs. USB vs. CC Mode

Thanks vinark, but dawbench is not so useful for me. I'm only interested in the actual hardware latency, not the performance of vst plugins and the cpu. Also I don't consider it really cross platform if there is no Linux.

  • vinark
  • DAW Mastermind
  • Offline
  • Registered: 2009-06-05
  • Posts: 2,647

Re: Latency tests with the Fireface UCX and Firewire vs. USB vs. CC Mode

Well then you did a good test yourself. I´m interested in what it tells you/why you need to know?

Vincent, Amsterdam
https://soundcloud.com/thesecretworld
Babyface pro fs, HDSP9652+ADI-8AE, HDSP9632
  • mxa
  • ADAT user
  • Offline
  • Registered: 2012-11-25
  • Posts: 20

Re: Latency tests with the Fireface UCX and Firewire vs. USB vs. CC Mode

For use in a setup where the computer is an effects processor for live music or the instrument is the computer latency is key. The latency arguably should be below 10 ms for an instrument like a guitar or below 5 for voice. Keep in mind that you need to subtract the 20ms added from the software in the numbers above and also that it is measured round trip - for a software instrument only the latency added by the d/a conversion is relevant, while for an effect processor from a (microphone) input both a/d, d/a and the time the actual effect processing needs are accumulating. http://en.wikipedia.org/wiki/Latency_%2 … erformance