Missing Metrics - Latency!

[drumphil]

While I think audio fidelity is well covered in the measurements reported in reviews here, as someone who works in the field of audio production, there is one thing missing that does actually matter when you're actually working with audio, beyond fidelity..

Latency!

It would be awesome if all reviews involving ADC, DAC, or just digital processing included latency figures. Because when you're dealing with real time audio as a performer, latency is critical. It doesn't matter how good your gear sounds if there is a 50ms delay somewhere in the signal chain between the performance, and the artist hearing it.

A practical example; I play drums, and I have electronic drum kits that I use to trigger virtual drum software. I have audio interfaces that are capable of low latency, but in that category of equipment, with the exception of some very recent and commonly expensive options, it is hard to get a really good quality headphone output.

I have an SMSL C200. Sounds amazing, better DAC than my audio interfaces, and much better headphone output quality, especially into lower impeadance headphones, or IEM's, but I would love to know for certain, if I attach it to the SPDIF output on one of my audio interfaces, am I going to be adding a bunch of extra latency? Perhaps it will actually perform better than the DAC chips in my audio interface.

Timing is critical to musical performance, and if piece of equipment compromises that by delaying what I hear, it is not a viable option. No if's but's or maybe's about it.

If the only audio interfaces/DAC's in the world that had really good latency performance had a noise floor around -70db, I'd be using that over anything else at any superior level of fidelity.

Would save a heap of time having these figures available, as it allows for a neat dividing line between everything that has good enough latency performance (which is of course subjective), and everything that doesn't so I don't have to waste any time considering those options.

So that's my request to you Amir. Latency figures. Doesn't matter much if you're just listening to music or audio, but makes a big difference if you're creating it, and this information is otherwise frequently nearly impossible to come by, outside of equipment specifically designed for audio production, and even there accurate figures can be hard to come by.

To me, this is like headphone output impedance, which is actually really important, but until recently a mystery with most equipment. Would be great to see latency added to the list of things that are no longer a mystery if we have the good fortune to have a piece of equipment we're considering reviewed by Amir.

 

[kemmler3D]

Good point, I would just play devil's advocate on this a little and point out that latency in practice will depend on the OS and driver you're using too... One device might have many latency values depending on the setup. I expect Amir will not want to test every connection and computer connection for this... But someone should!

 

[drumphil]

Good point, I would just play devil's advocate on this a little and point out that latency in practice will depend on the OS and driver you're using too... One device might have many latency values depending on the setup. I expect Amir will not want to test every connection and computer connection for this... But someone should!

Typicaly for a given buffer size (asio driver buffer size on PC, core audio on Mac) the figures don't change with driver or OS versions, with a few exceptions.

Generally what you get from an specific audio interface at a given buffer size on a given platform, remains the same.

Of course it's much more simple for things like SPDIF to analog and vice versa in DAC/ADC equipment. Just a single figure per sample rate.

I would love to be able to say that I have seen an audio interface with high latency at a given buffer size, reduce that latency with a newer driver version, making old figures obsolete. I can't think of an example when I've seen that.

I'm pretty sure my M-Audio Audiophile 2496 has the same latency at all buffer sizes in the Windows 11 system it's running in now, as it did in a Windows XP system years ago.

It's actually quite hilarious. I'm running that card in an i7 based Windows 11 system now. 4.5ms round trip latency at a 64 sample ASIO buffer size. The sound on sound review of that card was done using a Pentium II 450 system, running Windows 98, in 2001.

How many bits of gear have that kind of longevity, and still competitive performance!

 

[TK_IIVI]

Not sure if this is of any help, but in case it is:

Audio Interface - Low Latency Performance Data Base - Gearspace.com

Hey All, I am sure all of us have seen posts over the years where end users have purchased a new audio interface and have experienced varying degrees o

gearspace.com

 

[drumphil]

Not sure if this is of any help, but in case it is:

Audio Interface - Low Latency Performance Data Base - Gearspace.com

Hey All, I am sure all of us have seen posts over the years where end users have purchased a new audio interface and have experienced varying degrees o

gearspace.com

Because audio interfaces are frequently used by musicians, that sort of thing exists, but it is much harder to find that kind of information about stand alone ADC and DAC units and signal processors.

 

[RexrothPigeon]

Yeah, that's more of a gearspace topic. this forum is mainly aimed at home listeners, not musicians.

 

[drumphil]

Yeah, that's more of a gearspace topic. this forum is mainly aimed at home listeners, not musicians.

I would like to think it's focused on audio science, that has implications far beyond just media consumption for home entertainment.

For those involved in music production, latency is an important metric when evaluating equipment, frequently more important than just about everything else.

 

[Fongus]

Audio latency matters in gaming too. Especially for wireless audio.

 

[john61ct]

And many of us use our music system for watching shows and films.

But then at least we can delay the video side as needed.

True for "live" monitoring and gaming, that's just not possible.

 

[drumphil]

But then at least we can delay the video side as needed.

This frequently falls apart too, if reported latency is incorrect, or any such reporting system is just plain absent, like with an external SPDIF DAC. No way for the rest of the playback system to know how much delay is added after the SPDIF output.

Manual compensation for such things would be much easier if we had latency measurements in reviews of audio equipment.

 

[john61ct]

Manual compensation for such things would be much easier if we had latency measurements in reviews of audio equipment.

Unfortunately each step along the way adds its own

often VERY variable depending on that room and all the gear in the chain and how it's used.

I agree that latency "should be" measured for components where it ISN'T so variable

IF those doing the measuring also recognise the need.

But in the end, for accuracy in fixing lip sync you can't just add up a stack of numbers, you need to measure the end result, or just use trial and error.

 

[drumphil]

I agree that latency "should be" measured for components where it ISN'T so variable

Just basic ADC/DAC conversion latency for stand alone ADC/DAC units and ASIO/Core audio I/O latency at a few common buffer sizes for audio interfaces/sound cards would go a long way. That sort of equipment is much more likely to be relevant in an audio production system than something like a home theatre surround amp, or bluetooth headphones.

 

[phofman]

IMHO latency of the OS stack will be larger by orders of magnitude than latency of the actual device. It's quite difficult to measure latency of a device alone via USB or SPDIF input. Hardware itself has typically very short buffers inside. Latency of DAC/ADC chips (i.e. of their filters) is typically listed in their datasheets. IMHO if people experience latency issues, it's almost all caused by buffers in software before the actual host interface (USB controller DMA side, SPDIF transmitter for DAC/receiver for ADC), not by the DAC/ADC hardware.

 

[john61ct]

Still, what I wrote applies to all that, whether music production or recreational listening.

With many DSP tools running on a computer, you have control over the buffer size, based on what you think your effect / filters will need.

For mixing and masterin,g latency is not an issue, only live monitoring and gaming.

 

[drumphil]

IMHO latency of the OS stack will be larger by orders of magnitude than latency of the actual device.

Not necessarily true, especially with devices optimized for really low latency operation.

For instance:

Purely buffer size round trip latency with no real world considerations, for a 64 sample buffer size at 44.1kHz is 2.9ms.

An RME RayDAT at 44.1kHz with an ASIO buffer size of 64 samples, has a digital to digital round trip latency of 3.742ms.

The round trip with a very low latency converter like the RMA ADI-8 QS goes up by 0.49ms to 4.232ms. But if you were to use something like the Behringer ADA8000, the round trip would increase by around 3ms to approx 6.7ms.

That's throwing away a good chunk of the latency advantage that costs a lot of money to get using RME audio interfaces.

https://www.tonstudio-forum.de/blog/entry/105-rtl-comparison-of-different-rme-products-en-de/

And before anyone say these figures are all too small for it to matter, that's not the only latency I'm dealing with. My e-drums have their own latency too, takes time between when I hit a drum pad and when the virtual drum software gets that information. If I'm doing this on stage there is extra latency with the sound running through a digital mixer before it gets to me, and more if I'm using digital wireless in ear monitors. It all adds up.

 

[Redacted]

For home listeners, many only find out about latency when using Bluetooth 'phones (horrible anyway) when they listen to TV through their stereos, and get lip synch issues. Often not fixable in a lot of kit by applying relative delays between video and audio.

Similarly, 'modern' musicians going from analogue gear to digital, like with my own guitar rig - you get a digital link from guitar, then a modeller/effects pedal, maybe an amp fronted by non-switch-outable effects, then maybe a digital wireless IEM to listen to your cool sounds, and you find you have up to 1/4 sec delay from plucking a string to hearing it. Impossible to use. I've had to choose carefully (on a budget) gear that plays OK (Valeton GP5 [in or out of FX loop], Lekato 5.8GHz guitar link, an analogue amp, albeit hydrid 'cos of reverb) and that is close to the limit for live sound. Recording a track at a time onto PC/Audacity through a cheap Behringer ADC is OK too. I appreciate how master clocks have to to be used for complex, multichannel recordings, but that's way beyond my scope!

Long story short - published figures for latency on any piece of gear, especially if it involves DAC or ADC are useful to know. I bought my most recent cheapo guitar bits based on checking published latency figures - once bitten, twice shy....

 

[phofman]

An RME RayDAT at 44.1kHz with an ASIO buffer size of 64 samples, has a digital to digital round trip latency of 3.742ms.

A round trip means that it is 1.9 ms in one direction. 64 samples buffer at 44.1kHz is approx. 1.5ms for the ASIO buffer. The remaining 0.4ms is shared by the USB core driver (which assembles data for each USB function into the USB packet for USB-controller to read over DMA, then the USB controller latency, and latency of the actual device (which is what I talked about). So the latency of the actual device may be at 0.2ms - which means the OS/host side latency being almost an order of magnitude larger. That's what I was saying.

 

[drumphil]

A round trip means that it is 1.9 ms in one direction. 64 samples buffer at 44.1kHz is approx. 1.5ms for the ASIO buffer. The remaining 0.4ms is shared by the USB core driver (which assembles data for each USB function into the USB packet for USB-controller to read over DMA, then the USB controller latency, and latency of the actual device (which is what I talked about). So the latency of the actual device may be at 0.2ms - which means the OS/host side latency being almost an order of magnitude larger. That's what I was saying.

Ah, I thought you were including ADC/DAC latency as part of "device latency", thinking you meant that audio "device latency" was largely irrelevant due to the vast majority (orders or magnitude more) of the latency being in the OS audio stack, which doesn't make sense if the ADC/DAC is part of that comparison.

Device is a loose term.

 

[Ropeburn]

Typicaly for a given buffer size (asio driver buffer size on PC, core audio on Mac) the figures don't change with driver or OS versions, with a few exceptions.

Generally what you get from an specific audio interface at a given buffer size on a given platform, remains the same.

Of course it's much more simple for things like SPDIF to analog and vice versa in DAC/ADC equipment. Just a single figure per sample rate.

I would love to be able to say that I have seen an audio interface with high latency at a given buffer size, reduce that latency with a newer driver version, making old figures obsolete. I can't think of an example when I've seen that.

I'm pretty sure my M-Audio Audiophile 2496 has the same latency at all buffer sizes in the Windows 11 system it's running in now, as it did in a Windows XP system years ago.

It's actually quite hilarious. I'm running that card in an i7 based Windows 11 system now. 4.5ms round trip latency at a 64 sample ASIO buffer size. The sound on sound review of that card was done using a Pentium II 450 system, running Windows 98, in 2001.

How many bits of gear have that kind of longevity, and still competitive performance!

You must have one of the last mainboards with PCI (not e-). Basically the same with my old i7 (1st gen) PC and Delta 1010. Those mainboards were pretty much the last to have PCI slots.

It's kinda weird that 20 year old PCs with PCI audio interfaces had ultra low latency, stable to full CPU load, as standard. And then the technology changed without necessarily improving: it's actually harder to find USB interfaces today offering the same performance and rock solid stability.

 

[drumphil]

You must have one of the last mainboards with PCI (not e-). Basically the same with my old i7 (1st gen) PC and Delta 1010. Those mainboards were pretty much the last to have PCI slots.

It's kinda weird that 20 year old PCs with PCI audio interfaces had ultra low latency, stable to full CPU load, as standard. And then the technology changed without necessarily improving: it's actually harder to find USB interfaces today offering the same performance and rock solid stability.

Z77 chipset (Asus P8Z77-M Motherboard with i7 3770k cpu at 4.6gHz), which was the last intel chipset with native PCI support. I use that system for edrum stuff. Lol, that case originally had a pentium 4 in it, which was still newer than the Audiophile 2496.

But I have a crazier system than that! An Asus Pro B550-C CSM motherboard, which has a PCIE to PCI bridge chip, with a Ryzen 7 5800X, that I built because I got an RME HDSP 9652 for $200 (australian dollars).