I asked the question about impossibility to set buffer at 4096 just out of curiosity. Why other devices alowing me to set this but not BF Pro?
But my main question is, why the percieved latency and especially performance (lowest possible buffer without clicks and pops) are the same in low-end Behringer Xenyx USB and Babyface Pro? Is it something wrong with my Babyface unit or my setup, or just in that kind of circumstances I decribed in first post the difference doesn’t matter? As I said, before this test I’ve just expected any difference, but there was no percived difference. I don't understand that.

If your not recording and using real time effects as you record audio or play vst instruments from a controller then you probably won’t notice differences in what latency is set at as this is what really tests it. No idea why there is no option for 4096 or why it matters. I suspect in real world latency testing such as playing vsti’s from a controller or real time effects while recording would show up a significant difference.

I am changing buffer size in Digital Performer Audio Device settings. Why this comparison is useless? I’m not measuring latency by using special tools, I am not doing any scientific research. I just want to understand the latency/performance topic in practice and why my results are like I described in first post. And if this is a normal behaviour, or I can fix something.

I have little experience in this field, because in the past I didn’t need that. And my very amateur observations are: when I set 1024 buffer size, the latency is very clear audible, at 512 very little and under that value unnoticeable. At 32 and 16 buffer I hear distortions. If latency and distortion occur with the same buffer values both on Behringer and Babyface, they seems to be similar in performance. And the question is WHY and if this is normal? I suspect it’s impossible, that Behringer is as good as BF, so my conclusion is, that I don’t undestand something here and I would like to now what.

I’m not quite sure, if I understood you. Do you mean that I should install ASIO driver on my Macbook Pro to percieve the performance difference between Behringer and Babyface?

Sry, overlooked you have Apple, ASIO is only for Windows.
In Apple you set buffers in the application as far as I understood it from reading various posts.

I would think then like you, that buffers,
- be it inside of the ASIO driver under Windows
- or partially implemented as a safety buffer / in the application under Mac OS,
finally are for the same purpose and can be measured and compared.

Channel count has nothing to do with latency over USB/FW/TB.

@Del_Gesu: perhaps the following table will help you, in which I have recorded the RTL (round trip latency) of various RME recording interfaces at 44.1 kHz (single speed).

Not only for recording interfaces alone but also for two different use cases with connected devices through ADAT and MADI:
- UFX in standalone mode as preamp in front of a PCIe based RayDAT card
- One Octamic XTC connected to HDSPe MADI FX through MADI
There you have to add the time for A/D and D/A of this particular device additionally.
The transport time over ADAT or MADI is so low, you can ignore it silently without getting too inaccurate values.

In terms of RTL (round trip time) is the time for
- A/D conversion (e.g. from Mic)
- Time for transport over USB/FW/TB/PCIe/... to the PC and back
- D/A conversion (e.g. to headphones)

This is what you measure when using the external tool that MC mentioned in post #12.

One more comment: RME codes the RTL (round trip latency) very accurate into their drivers (what is not the case for every vendor), therefore you can get very reasonable values when looking to input/output latency e.g. in the driver section of Cubase like I did.

Important findings can be derived from the table:
1. The latency times of all RME recording solutions are excellent, regardless of whether they are based on Firewire, USB, PCIe or Thunderbolt (external PCIe) and also regardless of the number of transmitted channels
2. Time-critical applications (eg playing on virtual instruments, VSTi) require an RTL of less than 10ms and this is what you achieve with all RME recording solutions with ASIO buffer sizes <= 128 samples at single speed.

If you read the reviews of recording interfaces from other manufacturers carefully, you will find that the RTL of recording interfaces from other manufacturers with the same ASIO buffer size and the same sample rate is significantly worse, sometimes even twice as high.

RME has a couple of decisive advantages here
- Excellent know-how to program efficient drivers that meet the specifications
- Solid design decision to handle the communication to the computer via the FPGA and not to use 3rd party chips
- Solid knowledge of FPGA programming, USB, FW, Thunderbolt communication to be able to program according to the specification

Other things that can be derived from this. RME offers you long term driver support and a high investment protection when purchasing a device, because here everything essential can be fixed via firmware upgrades of the FPGA.

Devices of other vendors with cheaper designs (consumer area) without FPGA based design or where the communication to the computer is implemented by 3rd party chips, you can only exchange the device in case of failure and many manufacturers support their consumer products maybe 3-5 years and you sometimes only get updates rarely.

With RME you get even after many years (some products are already almost 20 years) even free function updates for features that were not even intended for the device from the beginning. Best example: support for the ARC USB remote control for devices for which a remote unit was not even planned at that time.

When comparing devices, you will also notice that many other manufacturers offer devices with USB3 and Thunderbolt, although the devices only have a comparatively very small number of I / O ports.
My guess is that this is an attempt to cover up fundamental deficits in driver programming (efficiency, latency).
This is complete nonsense, because it only makes the devices more expensive. For example, RME can transmit up to 68 IN / OUT channels via USB2 (see MADIface Pro product).
And as a customer, in addition to the price advantage, you have the additional advantage that USB2 is supported by almost all computers, whether desktop or laptop and that if there are few USB3 interfaces available, these can then be used for backups. This saves you having to use a USB3 hub.

To make a rather long story short and maybe save you a lot of measuring work.
Get an RME recording interface and you will get excellent quality and performance in all areas.

RME is really ahead in all areas (drivers, features, technology (SteadyClock, MADI features, ...), user interface (TotalMix, ...). ), user interface (TotalMix FX, TotalMix Remote), detailed and well-structured manual in 2 languages (de,en), Long Term Firmware/Driver Support, FPGA based design, Transparent Mic Preamps with very good gain range and high level stability, very good technical data for all inputs and outputs, Excellent support especially via the RME User Forum, Direct contact with developers, Support for optional ARC USB, free additional software (DIGICheck), ..

I have now dealt with this and similar topics for a long time, RME is simply the best choice. You get sophisticated and innovative studio technology at a reasonable and fair price.

See also the CrispyChips comment on the quality of the very nice transparent RME Mic preamps.
Also very interesting and important to know and consider for a purchase decision:
https://forum.rme-audio.de/viewtopic.ph … 99#p180399

One last thing, the Babyface Pro FS got a facelift, approx 2y ago. It has currently the fastest converters built-in.
But pls note, converter latency is much lower compared to the transport over USB/FW/TB/...
Nevertheless it reduces the overall RTL and has an even lower "near realtime latency" when routing audio through the digital matrix on the recording interface itself, e.g. from mic input to headphones.

EDIT: and to your last question why you get sometimes audio dropouts.

With a higher channel count more data needs to be transported and processed by the CPU.
The lower the ASIO buffersize is, the more often the CPU gets interrupts from the driver to perform I/O for this device.
As a result of this CPU load / interrupt load increases.
It depends now on a couple of factors, how well and how efficient a computer is able to handle such an audio load.
Bad mainboard designs / drivers that occupy CPU cores for too long can block those cores from processing audio related tasks, which have real-time demands. If you have a very old PC also CPU performance can be an issue.

If RME drivers are well written and can deliver audio with a much lower latency this means for the computer that he also has to react more quickly and to process audio in time. This needs lower DPC latencies. Here some systems can have an issue.

This is one of the reasons why it can be beneficial to purchase a turnkey system for audio, where experts in this area take selected components, which ensure, that all PC components can work efficient without high DPC latency where drivers can block CPU cores for too long, which is counter productive for processing audio in time.

Getting CPUs with a high single thread performance is usually beneficial as well, as then one CPU core (/thread) can process one thread very quickly to save time and compute audio in time (e.g. if you tend to use many inserts in a track or if you have CPU hungry VSTi).

So .. your performance / latency questions goes even further
1st of all how you work with your PC (for pure recording the ASIO buffersize can be set to max without any disadvantage).
And optimum results you can only achieve if all components - which have a performance relationship to each other -
are optimized or at least have a certain "quality" in sensitive areas (driver, DPC latencies, ...) which are important for audio processing.

Like in Forula-1, its not only Horse Power of the formula 1 car. Its excellence in all important areas (including the whole team behind) and even depends on the weather (which is an unknown factor like perhaps which kind of DAW project you have or which programs work additionally in the background of your PC).

To sum up

Sorry, it's a bit longer, but I hope I could show some interesting connections.

You don't have to worry about latencies with RME products, rather the opposite.

If something should not run so smoothly (e.g. audio dropouts with very small ASIO buffersizes), then you should really think about your computer. Sometimes it's just a few "small things" in the area of BIOS/driver/power settings of PC or graphics card, sometimes it's the mainboard itself (design/BIOS/chipset) and you don't get DPC downsized significantly.