Minimizing latency with macOS

Latency is the time taken for a certain action or process to occur.  For Lynx Studio Technology converters, we have done everything possible to minimize the amount of latency through the converter.  Latency comes from the DAW software buffer size (usually 32 samples to 2048 samples or higher), internal operating system buffers, and from the hardware converter group delay. 

Latency is usually specified in samples, but can be converted to milliseconds by dividing the samples by the sample rate.  For instance, a 32 sample buffer at 44.1kHz will be at least 0.73ms just for the output buffer and 0.73ms for the input buffer. There is also the internal group delay of the converters and safety offsets for the operating system, so the minimum round trip latency at 44.1kHz is going to be higher than just adding the input sample buffer and output sample buffer together.

There are three controls in NControl that will affect overall latency on macOS: Payload Size, Input Safety Offset and Output Safety Offset.  Normally all three values are set to Auto. 

When set to Auto, the values are set very conservatively to keep from dropping out audio with massive CPU loads with high channel counts and high sample rates.  If you are not using a lot of CPU for plugins, it should be fine to turn the values down. Remember that these values are sample rate specific, so running at 44.1kHz will be one set of values, and running at 96kHz will require larger values.

You can fine tune this to your specific setup in NControl.  To really dial down the latency, set the Payload Size to 8 or 16 (8 will be lower, but more chance of dropouts) and set the Input Safety Offset to 16 and the Output Safety Offset to 32. That should be about 4 ms at 44.1kHz (about where it is when everything is at Auto). Starting turning down the Output Safety Offset until the audio breaks up.  You can only change the value when audio is not streaming, so you have to exit your DAW software or change to a different interface while changing the value. Once you have it at the lowest value for the output where audio does not dropout, you can start turning down the Input Safety Offset.

The theoretical limit would be 32 samples + 18 samples (input group delay) + 32 samples + 16 samples (output group delay) or 2.3ms round trip, so around 2.5ms is really as low as is possible and still not dropout audio (minimum values for Payload Size, Input and Output Safety offset). The 18 and 16 are the converter group delay for input and output respectively (how much time it takes to get through the actual chip) and of course the 32 samples are the lowest buffer size possible.