3

I'm trying to determine Round-Trip Latency of realtek built-in audio interface using CEntrance Latency Test Utility:

enter image description here

To do this it's necessary to connect audio interface output to input using a short 2.5mm audio cable: enter image description here

The problem is that it gives inconsistent results (15 samples):

5.44, 5.44, 5.44, 48.50, 5.44, 5.44, 5.44, 39.57, 35.76, 5.44, 49.93, 5.44, 5.44, 5.44, 35.19, ... (in milliseconds)

I've tried various buffer sizes, input sensitivity levels in the Centrance tool to no avail.

I've also measured latencies for the system using DPC latency checker and Latency Mon:

enter image description here

enter image description here

It appears that there is no apparent driver stalls or hiccups in the system.

How to fix this? How to find the source of inconsistency? Is it the tool itself? (is there a better tool?)

Improve this question
3

I've managed to get consistent results now!

~7ms with 128 samples, ~4ms with 64 samples @ 44100 Hz.

I had Audio Out connected into Audio Mic Input port.

Now it is connected from Audio Out into Audio Stereo Input.

(i'm not sure why that made the difference, or if the measurements are reliable)

Update: Out of curiosity, to make sure that CEntrace tool is reliable. I made a small C# .NET Winforms app which does the same (using NAudio library which has Asio audio driver support):

enter image description here

Here the latency is ~4.5ms with 64 samples @ 44100 Hz and ~7.5ms with 128 samples - which is in the same ballpark figure as provided by CEntrance tool.

Improve this answer
1

What you interpret as inconsistency is just the normal system behaviour. I would say your results are pretty consistent, considering the normal variance that can occur in a multitasking system running multiple applications. Even if you have a system running only your audio applications as end user applications, normal system functioning (memory management, process scheduling, and other system tasks) are going to produce seemingly random variation in processing time of other tasks.

That's why you have dropouts: random moments in time where several factors congregate in the worst possible way to impede the audio application to clear a buffer within the required time. If all was constant, there would not be dropouts, either you would be able to process a certain buffer size in due time (and all is well) or you wouldn't (and have no sound at all).

So what you can manage is the probability to have dropouts, and what the tools are telling you is that, in the current configuration, that probablity is very low (perhaps close to zero, but that may change once you start using plugins with high CPU demand...).

It would take a dedicated system with a real time operating system to have (more) precisely constant latency.

Improve this answer
3
  • Performance hiccups in the order of 40-50ms is not normal system behavior. The measurements made by LatencyMon and DPC Latency Checker do not indicate any of this happening. In fact, the highest system latency measured is 0.3ms, while the usual is around 0.05-0.10ms. If the system had regular hiccups of 30-50ms, It would be noticeable even from regular system use (especially when playing games). In modern computing milliseconds of latency is a literal eternity. For the reference, if it took 40ms to render a frame in a video game, that would result in lowly 25 frames per second. – JBeurer Jun 28 '16 at 4:01
  • A simple tetris/mario game written in C/C++ would run at consistent 4000-5000 frames per second on this system (with all the multitasking OS stuff running at the same time), that's 0.25ms-0.2ms per frame utilizing only one out of 4 CPU cores (and even then in not particularly efficient manner) – JBeurer Jun 28 '16 at 4:15
  • I see what you mean I guess my sensibility is conditioned by my own machine. I do get hiccups of 50 ms and have occasional dropouts in the DAW when using heavy plugins. So I suppose my answer may have a theoretical soundness but is not applicable to the situation. I'm considering deleting it, but I think it would interesting that you included your comments above in your own answer (or original question) as it helps to understand the problem. – José David Jun 28 '16 at 14:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.