I continued to look into audio latency and focused on audio only measurements this time.
In essence I sent a test signal in various sample rates to a USB to coax converter, split the coax output into two (an RCA Y splitter is good enough for this test), feeding both a Focusrite Scarlett 6i6 2nd Gen and one of several DACs. The given DAC's analog output also went into the 6i6. I then recorded a stereo audio file, the left channel containing the left digital channel and the right channel containing the left analog channel from the connected DAC.
With a little script I looked for the peaks in both channels (to be insensitive to filter ringing) and the number of samples between the first peak of each channel tells me the latency between the digital signal being received and the corresponding analog audio being generated.
One caveat here is that I can't guarantee that there isn't some buffering at play in the 6i6, or that the buffering is the same for digital and analog inputs, or that it doesn't vary with the sample rate, so while the absolute numbers may not be entirely accurate, the differences between DACs should be accurate. I set the 6i6's clock source to S/PDIF instead of internal.
I could have used the 6i6 itself as a USB to coax converter but for some reason it couldn't lock on to its own S/PDIF signal for sample rates greater than 96 kHz, so I used a Topping D10 instead, which worked fine. An added benefit is that the D10 also has a TOSLINK out and an easy to read sample rate indicator. I don't think I have a way to verify that its coax and TOSLINK out are completely in sync, but the results I got suggest they probably are. This way I could also compare a DAC's latency when using its coax input vs. its optical input, and not surprisingly there isn't really a difference with the DACs I tested.
The bottom line is that there are two clusters:
- Cluster A: Modi Multibit, Bifrost 2 and Yggdrasil OG
- Cluster B: Bifrost 2 (NOS mode), Topping E30, iFi micro iDSD (only coax tested), Reiyin DA-Super, Focusrite Scarlett 6i6
For the iFi micro iDSD I used firmware 5.2. The Bifrost 2 was first tested with the firmware update Schiit released in September of 2019 or so, but I also tested it with the new firmware with NOS mode released last month. I'm not sure which filter was active for the Topping E30.
| Sample rate | Cluster | Delay (samples) | Delay (time) |
| 44.1 kHz | A | 142-147 | 3.220-3.333 ms |
| B | 9-39 | 0.204-0.884 ms |
| 48 kHz | A | 141-147 | 2.938-3.063 ms |
| B | 8-38 | 0.167-0.792 ms |
| 88.2 kHz | A | 54-71 | 0.612-0.805 ms |
| B | 12-38 | 0.136-0.431 ms |
| 96 kHz | A | 57-75 | 0.594-0.781 ms |
| B | 12-42 | 0.125-0.438 ms |
| 176.4 kHz | A | 23-62 | 0.130-0.351 ms |
| B | 4-44 | 0.023-0.249 ms |
| 192 kHz | A | 28-63 | 0.146-0.328 ms |
| B | 4-40 | 0.021-0.208 ms |
The differences between those clusters were most pronounced for 1x sample rates (44.1 and 48 kHz) where cluster A had a little over 3 ms of latency while cluster B was already below 1 ms. At 88.2 kHz and up all DACs had less than 1 ms of latency, but the clusters where still separate for 2x sample rates. For 4x sample rates (176.4 and 192 kHz) the clusters had some overlap.
The iFi micro iDSD with its "bit-perfect" filter was consistently the fastest for each sample rate and for 1x and 2x sample rates the slowest when using its standard or minimum phase filters (similar latencies for both).
Beyond 96 kHz, Yggdrasil and Modi Multibit somehow have even less latency than Bifrost 2 in NOS mode while Modi Multibit also had a lower latency than the iFi micro iDSD with its standard or minimum phase filters, hence the overlap for those clusters. While Modi Multibit is effectively in NOS mode for 176.4 and 192 kHz, Yggdrasil should still be upsampling and thus have higher latency, or so I would have thought.
A brief test of using Mutec MC-1.1+ DDC in between the D10 and Yggdrasil seems to have added up to 10 samples of delay, but for one sample rate was actually 3 samples ahead of the direct route, so this is inconclusive, although it doesn't seem to be a big concern. Latency of DDCs would be easier to test with an audio interface that has two digital inputs so I can replace the analog channel with an indirect digital channel, but the 6i6's two digital input channels are invariably part of the same stereo S/PDIF stream.
If any DAC had been able to register its analog signal before the digital signal, we would have had proof that the S/PDIF input of the 6i6 is delayed compared to the analog inputs, but that was not the case, at least when only looking at the peaks (I should try and determine the noise floor so I can check for recorded pre-ringing before the first digital sample is received). It's clear that filters are a major contributor to latency because those DACs that can turn them off produce an analog signal much faster.
It's possible that the latencies measured here are close to the actual total latencies for each DAC.
If so, all DACs having less than 3.4 ms of latency tells me that any audio/video synchronization concerns appear to come from somewhere else in the chain. I also repeated some of my slow motion video tests and this time the Topping E30 had less latency than the Bifrost 2. My previous test method (recording 240 FPS video on a phone) appears quite imprecise, at least with the equipment used. A frame rate closer to 1000 FPS and an external microphone input would help.
Now what casts doubt on my results is what Jason has previously said about latency:
But maybe that was for USB only which may well have some additional potential for latency due to various buffers before the audio even hits the DAC.
Good point, actually. About 1.6m was probably the furthest, so less than 5 ms, I'd say. And the closest was the phone between headphones, so definitely a contributor.
One assumption I made is that analog equipment (amplifiers, analog cables, transducers) don't meaningfully contribute to latency, but I mixed a speaker setup and a headphone setup, so maybe there's more than distance that I failed to take into account.
!!!! The Nod of Knowledge! I have zero clue as to what you are speaking on but have no doubts it is highly scientific and thereby beyond my ability to understand, ergo, I shall give you the Nod of Knowledge as though I were hip to your jive!
