USB latency

[Dale-c]

Has latency of USB interfaced caught up with firewire in the newer devices? There used to be a lot of sub-$500 FW interfaces but now that Thunderbolt has replaced firewire, it seems the TB interfaces are all fairly expensive and the only stuff in the mid and lower ranged are USB.
Is that a step backwards or have newer USB-C devices gotten to be as low latency as the old FW devices?

[Sardi]

RME interfaces using USB deliver the same latency as TB. That’s also USB 2.

Drivers, drivers, drivers.

Not all manufacturers are the same. RME aren’t. necessarily the only ones.


Sent from my iPhone using Tapatalk

[Darryl Ramm]

Well done curated lists of interface latency measurements exist on the web, you can find all that with a Google search.

USB is not being replaced. And Thunderbolt still has limited availability on Windows PCs, even with USB 4's optional inclusion of Thunderbolt 3.

[Dale-c]

Originally Posted by Darryl Ramm

Well done curated lists of interface latency measurements exist on the web, you can find all that with a Google search.

USB is not being replaced. And Thunderbolt still has limited availability on Windows PCs, even with USB 4's optional inclusion of Thunderbolt 3.

I have looked but I hadn't been able to find good specs.

It would seem that USB with the latency of firewire would be ideal now. I just picked up a firewire only MOTU Ultralite. I like it so far for $80 and, again, drivers are important. I have it running on my regular system with Catalina as well as my Pro Tools partition.

[JFreak]

USB 2.0 negotiate at 10 or 20 ms latency depending on features used by the drivers. Good luck going below that. Newer versions obviously faster, but the thing has not been designed for low latency to begin with.

[Darryl Ramm]

Originally Posted by Dale-c

I have looked but I hadn't been able to find good specs.

It would seem that USB with the latency of firewire would be ideal now. I just picked up a firewire only MOTU Ultralite. I like it so far for $80 and, again, drivers are important. I have it running on my regular system with Catalina as well as my Pro Tools partition.

Nit sure why Google did not turn it up, but the main latency database out there that folks refer to is results from DAWBench. People involved in that know what they are doing. Many other commentators on latency related stuff do not.

https://dawbench.com

and a widely used other link to Dawbench is https://gearspace.com/board/music-co...data-base.html Look for the link to the DAWbench PDF doc on that page, that link is kept up to date.

I don't see your interface reported in those results. But notice for what is reported that a Motu USB interfaces have lower RTL than the 828 Mk III Firewire interface. The devil us very much in the details with anything involving latency, watch out for ******** like "USB can't do low latency" or "you need thunderbolt" etc. Driver versions matter, OS versions can matter, many driver settings can matter, DAW version and setting matter, etc.

And note DAWbench uses (modest) workloads that the system runs while measuring latency, focus has historically been on Reaper. But importantly it's not a no-load latency measurement.

Much of the problem with latency masturbation on forums... it's not what somebody else can achieve in a theoretical world but what matters is what IO Buffer setting can you achieve for your DAW and computer and workload, since IO Buffer size will often dominate latencies.

You can use the same RTL Tool to measure latency used by DAWBench: https://oblique-audio.com/rtl-utility.php and see how you are doing for your setup vs. others. Dale - since you raised the whole latency concern, please do measure what you have and share here.

(want even more detailed breakdown of CoreAudio latency components I have a utility).

[Darryl Ramm]

Originally Posted by JFreak

USB 2.0 negotiate at 10 or 20 ms latency depending on features used by the drivers. Good luck going below that. Newer versions obviously faster, but the thing has not been designed for low latency to begin with.

Uh no. As already pointed out in this thread USB 2 *is* capable of low latencies, and what you get is highly dependent on the driver implementation. Certainly latencies well below 10ms or 20ms RTL (let alone 10ms or 20ms one way latency) are possible.

It takes a development team that knows what they are doing. And we've already pointed to RME being a leader there. And since my RME UFX+ does Thunderbolt, USB 3.1, and USB 2 and USB 2 class compliant I can report those RTL measurements. Or at least I would have Thunderbolt measurements if I could find my long Thunderbolt cable. Anyhow these are some no-load measurements (Pro Tools on this system routinely gets used for tracking a few channels in small sessions in Pro Tools at 32 samples with this interface vi USB 3 @ 96 kHz).

                          Measured RTL  @ 96 kHz, samples / ms 
                          ----------------------------------------------------
I/O Buffer size:          32 sample                  64 sample 
USB 3 (RME driver)        221 / 2.30 (173 / 1.8)*    285 / 2.97 (237 / 2.47)*
USB 2 (RME driver)        221 / 2.30 (173 / 1.8)*    285 / 2.97 (237 / 2.47)*
USB 2 Class compliant     --                         562 / 5.85

* Measurements in brackets were made with the RME driver set to use the short safety offset. Short safety offset mode may or may not be usable depending on load on the system/DAW etc. The other measurements not in brackets are made with the driver using the standard safety offset.These are actual RTL measurements made with RTL Utility using line level analog output to line level analog measured and checked with my own tools.

RME Fireface UFX+ Firmware v54
RME USB Driver version v3.27macOS Big Sur 11.6

The UFX+ allows you to switch between USB 3 or USB 2 or USB 2 class compliance modes, the Fireface USB Setting utility is used to check what USB mode the interface is in.

And whoa surprise surprise (it was a surprise to me) the USB 2 latencies are exactly equal to the USB 3 latencies. And yes that is real. Under the covers there are differences as well, USB 2 is always running to isochronous mode, USB 3 is defaulting to bulk mode, you can force the driver into isochronous mode (say if the interface is getting errors). Changing that mode made no difference to the latency measured.

And while YMMV as to how low an IO buffer size you can run with for your use, here is USB 2 and 3 showing impressive low RTL latency, that is far away from supposed USB 2 10ms or 20ms latency overhead.

[Sardi]

Originally Posted by Darryl Ramm

And while YMMV as to how low an IO buffer size you can run with for your use, here is USB 2 and 3 showing impressive low RTL latency, that is far away from supposed USB 2 10ms or 20ms latency overhead.

Very well said.


Sent from my iPhone using Tapatalk

[JFreak]

Right, I eat my hat. Nothing beats good testing, well done Darryl

[dominicperry]

Reaper reports 1.0ms in/1.0ms out with my RME Babyface Pro fs at 96KHz, with a 32 sample buffer. This is using USB2.0.

This is on an M1 Max MBP (in signature) using RME's latest 4.04 beta drivers, but IIRC I got the same figures for the official 3.27 driver. Both drivers are native AS, but the new beta drivers are moved out of the kernel, as per Apple's latest security model.

Dominic