ASIO Latency

[AlanH]

As I normally record live stuff, monitoring latency is not normally an issue for me.

But recently I have been helping out friends to set up their Macs for recording, and it frustrates me no end that they can plug my Focusrite Scarlett 2i4 into their MacBooks, spark up GarageBand and immediately get imperceptibly low monitoring latency (less than 5ms, I estimate).

In comparison, I have done everything I can think of with my Core i7 Windows 7 desktop PC to get the lowest possible latency in SAWStudio (and also Cubase and Ableton Live - it's obviously a general PC thing), and the lowest stable latency I can get (with the current ASIO driver set to 2ms/89 samples and SAW buffers 1) is an actual measured value of 26ms - which is borderline unusable.

Can anyone shed any light on this? Is it hardware thing or an operating system thing? If the former, maybe I should get a MacBook and put Windows on it!

As a dedicated PC user, it's really hard to accept that Apple have made an apparently much better job of this aspect of recording!

BTW, in case you were wondering why ASIO monitoring is important, the 'direct monitoring' system in the Scarlett 2i4 (and 2i2) is hopelessly quiet compared to the PC return signal - I have no idea why.

Cheers,

[Dave Labrecque]

BTW, in case you were wondering why ASIO monitoring is important, the 'direct monitoring' system in the Scarlett 2i4 (and 2i2) is hopelessly quiet compared to the PC return signal - I have no idea why.

Alan -- What do you mean by "hopelessly quiet?"

[Jeff Scott]

I use either a RME Raydat card or a Digiface as my PC interface and achieve about 9ms latency round trip at 2x64 buffer settings. It's not discernable to my ears and the Drumer I record with finds his Edrums "acceptable" at that rate. 1x64 is even better. That's on an older machine and/ or Laptop running Win XP.

At 26ms ...how is the Scarlett unit attaching to your computer? by USB 2.0?

[UpTilDawn]

As I normally record live stuff, monitoring latency is not normally an issue for me.

But recently I have been helping out friends to set up their Macs for recording, and it frustrates me no end that they can plug my Focusrite Scarlett 2i4 into their MacBooks, spark up GarageBand and immediately get imperceptibly low monitoring latency (less than 5ms, I estimate).

In comparison, I have done everything I can think of with my Core i7 Windows 7 desktop PC to get the lowest possible latency in SAWStudio (and also Cubase and Ableton Live - it's obviously a general PC thing), and the lowest stable latency I can get (with the current ASIO driver set to 2ms/89 samples and SAW buffers 1) is an actual measured value of 26ms - which is borderline unusable.

Can anyone shed any light on this? Is it hardware thing or an operating system thing? If the former, maybe I should get a MacBook and put Windows on it!

As a dedicated PC user, it's really hard to accept that Apple have made an apparently much better job of this aspect of recording!

BTW, in case you were wondering why ASIO monitoring is important, the 'direct monitoring' system in the Scarlett 2i4 (and 2i2) is hopelessly quiet compared to the PC return signal - I have no idea why.

Cheers,

This short article from the Presonus website offers an easy to understand explanation of how that latency you are experiencing is accumulated - and it's using a usb interface for the example: https://www.presonus.com/community/L...-Audio-Latency

It doesn't explain why there hardly any latency in your mac setup, but at least it's easy to see how the latency is generated on average...

Maybe the macs are setup to take advantage of the zero latency monitoring of the Focusrite, but your pc is somehow not using the zero latency feature?... just a thought.

[AlanH]

Thanks for the feedback that rolled in overnight.

Alan -- What do you mean by "hopelessly quiet?"

The Scarletts seem to use what the Presonus article (see below) describes as an 'old skool' approach to its direct monitoring, i.e. splitting the preamplified input before the A/D converter and mixing a proportion of it (fixed in the 2i2, variable in the 2i4) in with the output of the D/A coming from the PC. Setting the input gain to max out the input level, and looping it through the PC, with the 2i4 controls respectively fully CW and CCW, the sound coming back is much louder than the direct feed. This requires the mix knob to be adjusted a lot closer to the direct end than the PC end, leaving it to the overall headphone level control to get some 'beef' back into the headphone mix. Which it really can't do. Let's give them their due, though - this whole unit (preamps, digital conversions, 48V phantom power generation and headphone amplifier) is powered off the measly 2.5W available from the USB outlet.

I use either a RME Raydat card or a Digiface as my PC interface and achieve about 9ms latency round trip at 2x64 buffer settings.

That was my experience too. I also have a machine with a RayDAT card, and have used it for a few "PA-plus-record" gigs, and it works with very low latency. It is, of course, directly connected to the PCI bus, giving it immediate access to the DAW processing.

At 26ms ...how is the Scarlett unit attaching to your computer? by USB 2.0?

Yes, and that's the point here. It's the extra delay incurred by the USB gubbins that is causing the problem.

This short article from the Presonus website offers an easy to understand explanation of how that latency you are experiencing is accumulated

Thanks, that's a very good article, and sets out the points very clearly.

It doesn't explain why there hardly any latency in your mac setup

And that was the original point of my post: I have been resigned to the USB delays on the PC, and the consequent need for direct monitoring. It's just galling to find the Macs so much better in this respect.

My current everyday rig uses an RME FireFace UFX, which has excellent DSP-based internal monitoring with effects, making through-the-PC monitoring unnecessary. I only use the Scarlett with a notebook as a lightweight setup for ad hoc recordings.

Maybe the macs are setup to take advantage of the zero latency monitoring of the Focusrite

As far as I can tell, there is no such feature on the Scarlett; its 'direct monitor' function, as I describe above, is a just a simple analogue routing, outside the control of anything on the other end of the USB port.

I must conclude that Apple in their Macs have implemented some form of USB handling that minimises the delay in the USB Bus Clock Buffering stages (see the Presonus article), which seem to contribute the lion's share of the delay in a regular USB audio interface setup.

Next time I see my Mac-wielding friends, I will do a proper measurement of the GarageBand and Logic Pro live monitor delays.

Thanks again for the comments, guys!

Cheers,

[Cary B. Cornett]

Extra latency being imposed by going through USB instead of a PCI card raises the interesting question of why there are now fewer PCI card interfaces on the market. Are they being obsoleted in favor of USB? And will there be any reasonably-priced PCI card interfaces in the future? AFAIK, the available new options are all quite pricey.

As to the mechanism for zero latency monitoring, not having enough "gas" in the headphone gain after the necessary balance adjustments means the design engineers fell down on the job. It isn't about how much power to the cans (I'm sure there's plenty available), but the GAIN needed up front to get there.

[AlanH]

And will there be any reasonably-priced PCI card interfaces in the future?

I think it's unlikely, as it reflects a wholesale shift from tower PCs to laptops. Although I'm sure that USB will continue to dominate in the 'cheap-and-cheerful' market, I suspect that the more high-end interfaces will take advantage of newer communications technologies like Thunderbolt and AV-aware derivatives of Ethernet.

As far as I can see, there's nothing wrong with serial interfaces, as long as they have lots of bandwidth and can be used in a deterministic mode (i.e. with strict timing - which would also, I assume, reduce the need for buffering). I understand that the new generation of Ethernet-based standards (AVB?) are getting to grips with this.

It isn't about how much power to the cans (I'm sure there's plenty available), but the GAIN needed up front to get there.

Dead right - the headphone amp output, when fully driven, is perfectly adequate for anyone - other than a drummer, of course! I'm sorely tempted to hack the unit and add another 6-10dB of gain in the direct monitor feed, as the unit's well out of warranty!

For info, I have just been playing with my RME FireFace UFX using its USB interface, and the 'through-the-PC' monitoring delay, with a very conservative 128 sample ASIO buffer setting, is only 8ms, which is very comfortable to work with. This rather indicates that RME's drivers are more highly-tuned than the Focusrite ones. I seem to recall that Bob has always praised RME's performance on the drivers front, and now I understand why!

Cheers,

[Grekim]

Just curious how you are actually measuring the latency?

[dbarrow]

I had tried various Focusrite USB units and the drivers were crap. I couldn't even make a glitch-free recording on seeral different notebooks I tried. I have an RME Fireface USB and it seems to work flawlessly at 2x64. I have 2 Focustite OctoPre Dynamic units that either hook to the Fireface USB or a RayDat card on my other rig. I don't need the crappy Focusrite drivers for them, because they connect via LightPipe. The rock-solid RME drivers do most of the actual work.

[AlanH]

Just curious how you are actually measuring the latency?

This is my procedure:

1. Set up the DAW (doesn't have to be SAWStudio) to record a stereo track.
2. Connect a mic to the L channel of the interface under test.
3. Select 'tape-style' monitoring on the DAW.
4. Connect the interface L channel line out (carrying the microphone's monitoring signal) to the R channel line input of the interface.
5. Record a few impulsive sounds - tap a metal object etc.
6. Look at the captured file and measure the time delay between the L and R recorded impulses. All other things being equal, this is the monitoring latency.

It took me some while to realise that latency does not affect time alignment between tracks when overdubbing, because the interface reports its (perceived) latency to the DAW software, which offsets the sound in the captured file accordingly. But this does not help for 'live' input monitoring. And because the test described above captures signals on a single stereo file, the two channels are locked together in time.

Cheers,