USB vs TB Lateny: Is There Really a Difference?

[Tj Shredder]

https://gearspace.com/board/showpost.ph ... count=5092

At that link there are numbers I have no idea what they mean. What is RXC, CV, NCV, I/O, RTL??? There are no units to even get an idea what it means. The huge difference between the first and the last let me assume its complete meaningless bullshit...
The latency is mainly determined by buffer and block sizes and is independent of OS and speed of the connection (unless you run USB 1.1). There are differences in the amount of CPU an interface needs for their drivers, and it seems that this is measured here...
Already around the millenium we had audio interfaces on USB 1.1 which could deal with 8 channels of 44.1 kHz (Emagic EMI 2/6)...
20 years later we really don't need to worry about the speed of USB 3...

[Pictus]

At that link there are numbers I have no idea what they mean. What is RXC, CV, NCV, I/O, RTL??? There are no units to even get an idea what it means. The huge difference between the first and the last let me assume its complete meaningless bullshit...

Read :
http://www.original.dawbench.com/audio- ... atency.htm
http://www.original.dawbench.com/audio- ... tency2.htm
http://www.original.dawbench.com/audio- ... tency3.htm
IF you have any doubt, ask in https://gearspace.com/board/music-compu ... e-176.html

BTW, the PCIe/Thunderbolt have the potential to have less latency than
USB or FireWire, search for technical/white papers and you will understand why.

[Tj Shredder]

Reading those confirmed my suspicion. What has the number of RXC compressors running to do with latency? Its measuring overall performance which is something totally different… The only relation is, that you need bigger block sizes on slow computers which adds latency, but this has nothing to do with the connection to the interface…
I did read papers about Firewire vs. USB 2. That was 15 years ago and there was a fairly small advantage for firewire… Its 2022 and Thunderbolt and USB 3.2 are magnitudes faster… We have still CPU determining latency and less and less the interface connection…

[Pictus]

Reading those confirmed my suspicion. What has the number of RXC compressors running to do with latency? Its measuring overall performance which is something totally different… The only relation is, that you need bigger block sizes on slow computers which adds latency, but this has nothing to do with the connection to the interface…
I did read papers about Firewire vs. USB 2. That was 15 years ago and there was a fairly small advantage for firewire… Its 2022 and Thunderbolt and USB 3.2 are magnitudes faster… We have still CPU determining latency and less and less the interface connection…

As we can see at http://www.original.dawbench.com/audio- ... tency2.htm
RTL = round trip latency
RXC/CV/NCV = how many plug-ins you can run simultaneous = efficiency

Now lets look at some interesting results:

RME HDSPe AIO Pro PCIe has the best efficiency, not the best latency, but
the latency still lower than any USB.


The Quamtum Thunderbolt has the lower latency, but not the best efficiency.


The RME Fireface UFX+ has both Thunderbolt and USB and got interesting results
By Thunderbolt

By USB

Interesting, by using the USB the efficient is better, but as expected not the latency.
It is the same computer with the same OS, but different driver versions will give
different results, but no driver will make *any* USB to have less latency
than the RME HDSPe AIO Pro PCIe or the Quamtum Thunderbolt.

[golemus]

I use RME UFX+, comparing USB3 and TB connection, the latency under the same buffer setting the difference is approx. 0.7ms which means completely irrelevant as a consideration factor, unless of course, you are superhuman.

IMO if there is advantage with TB it is not the latency but another thing.

Have you ever compared TB vs. USB during very high CPU load (close to 100%)? I have all the time issue that there is crackling with high CPU loads and I have to change buffer from 128/256 to 512 to get it bearable. I was always wondering if this issue exists with TB interface as TB (AFAIK) does not need CPU power. Of course I have had only cheapish interfaces with worse drivers than RME

[sQeetz]

But you never know. Some people say Focusrite are bad for low latency (even on Mac), while others say they are very good for low latency.

Chiming in for the low latency camp on Mac. Focusrite Scarlett USB-C device. Smooth as butter on the rather old school M1 MacBook Pro 13. Either the 4i4 or the 2i2. 64 samples

[FapFilter]

That’s good to hear, but hopefully you mean 64 samples, as 64 ms wouldn’t be something that most people would regard as low latency

[sQeetz]

64 samples indeed... corrected. thx
To my feel and purpose it feels like instantaneous. Like there was no latency at all. I'm no professional by any means, but that's as good as I like it to be for my spontaneous music making endeavours....

[Bombadil]

On my RME Fireface 802, @64 samples, there was an advantage to the USB2 over the FW by about a millisecond, if memory serves.
On my M1 Mini, RTL is 4.9 ms @64 samples on USB. Can't complain about that.

[darkinners]

I use RME UFX+, comparing USB3 and TB connection, the latency under the same buffer setting the difference is approx. 0.7ms which means completely irrelevant as a consideration factor, unless of course, you are superhuman.

IMO if there is advantage with TB it is not the latency but another thing.

Have you ever compared TB vs. USB during very high CPU load (close to 100%)? I have all the time issue that there is crackling with high CPU loads and I have to change buffer from 128/256 to 512 to get it bearable. I was always wondering if this issue exists with TB interface as TB (AFAIK) does not need CPU power. Of course I have had only cheapish interfaces with worse drivers than RME

RME USB driver is flawless, even under extremely unreasonable CPU load, in a scenario where USB connection crackles, TB will too. I tested both.
But still, like you said, TB and the older Firewire always have an advantage that they have their own controller without too much dependency on CPU load to maintain good performance. However in my experience, RME really know what they are doing, as long as you don't plug your RME USB interface into the same USB controller that has multiple device sharing the resource already, I don't see any performance difference between USB3 and TB. except for a bit more latency with USB connection (less than 1ms)

[_leras]

I read all the stuff FireWire, tb usb when I had to replace a Mackie onyx that I couldn't get to install it's drivers.

I figured I'd go with RME on the basis of good drivers, long periods of support and lots of reports of improves sound. (Ufx+).

It's excellent. Low latency. Rock solid. Recorded audio sounds great played back audio is also excellent on monitors and headphones.

Yes it costs a bit more, but I'm confident it'll last many years and also be supported with new drivers. Can't go wrong in the long term. Very impressed with RME.

[Jim Roseberry]

It depends on just how low you want to push Round-Trip-Latency.
If you want to push RTL below 4ms, you'll want a Thunderbolt or PCIe audio interface.

Keep in mind that not all Thunderbolt audio interfaces offer super low RTL.
ie: The Fireface UFX+ is a rock-solid audio interface... but it won't allow RTL as low as the Presonus Quantum or Antelope TB units. The Apollo series are TB audio interaces, but they yield RTL equal to the best USB2 audio interfaces.

If you don't need onboard DSP for routing/mixing/loopback-recording, Presonus Quantum will yield 1ms total RTL at 96k using a 32-sample ASIO buffer size.
If you want onboard DSP for routing/mixing/loopback-recording, Antelope's TB series (like the Orion Studio Synergy Core) will also yield 1ms total RTL at 96k using 32-sample ASIO buffer size.

The best USB2 audio interfaces get down below 5ms total round-trip latency.
To get lower than ~4ms, you have to go Thunderbolt or PCIe.
The easiest way to think about Thunderbolt... is to consider it "external PCIe".

Working at high sample-rates and super small ASIO buffer sizes is extremely demanding on the machine.
To effectively do this... you have to be running a recent make well-optimized machine.