Focusrite Scarlett range refresh

[gimmeapill]

@bluebell: I have also found empirically that the 2i2 first gen performs better at higher frequencies(48 slightly better than 44.1 and 96k slightly better than 48k). No idea why, but if you want the lowest latency, better not stay at 44.1 khz and go straight to 96k (assuming the cpu, DAW and plugins behind can follow of course).

Looking at the benchmarks of the second gen, the best round trip latency figures are also achieved at 96k:
https://global.focusrite.com/scarlett-i ... technology

This is kind of counter intuitive, but good to know. Not sure if the other cards behave the same...

[raffguitar++]

Hi again,

@GraysonPeddie: Sorry about the Windows confusion. So, my studio laptop came with Windows 7 back when it was new and long before I discovered Linux. I installed all the applications that I wanted to use on it, and once I made sure everything was working correctly, I left the system as it is - meaning Windows has NEVER been upgraded on this machine. For one thing, this machine is not connected to the internet, so there is no risk of malware/viruses, etc, and also because I generally believe "If it ain't broke, don't fix it". Now, the Focusrite Windows driver refuses to install because Windows requires a very specific update to the entire OS (which I'm not going to do). For most users this won't be a problem, and Focusrite lists Windows 7 compatibility on their website...but this is yet another nail in the Windows coffin for me.

As far as the sound quality compared to older generations, I can't really say. I will say though that it does sound good so far. Interestingly, (..if my memory serves me correctly), it doesn't seem to sound any better than my old Emu 1616m (which also went up to 192k).

Anyway, I agree with you in that if the unit you're using now sounds great and works for you, there is no real reason to upgrade unless there are some SIGNIFICANT benefits to the newer model.


@gimmeapill: Thanks! I've been "lurking" in various Linux forums for sometime now, trying to learn from you guys. It's great to finally be here! As I mentioned, I'll try to report some more on this when my workload settles down a bit.



@bluebell: I'll have to experiment to see just how low I can get it; I've been using 128 (48k) without major isues, but as you say, when you've got a full session going with a lot of tracks it's a different story. I'm looking forward to my first real project with this.



Oh - there's one feature on the 2i4 that I think is cool: The Direct Monitor isn't a switch, it's a knob. So you can monitor the signal anywhere in between your direct input and what's coming back from the computer. This could come in handy if you're in a situation where the latency is getting really bad, you could dial-in some of your direct signal. Or even just for troubleshooting.

[GraysonPeddie]

I'm also a firm believer of "if it ain't broke, don't fix it" and I'd still keep Windows 7 even after Microsoft ends extended support by 2020. I'm a huge fan of Aero Glass and I did not like the look and feel of Windows 10, even if it's the way forward, so I went ahead with Linux after Windows 8 was released. I've been running Linux exclusively since November of 2012 and I did not look back. I simply did not want to pay for an OEM version of Windows 7 Home Premium after I build a new computer that's tied to a motherboard. And that is thanks to Valve for releasing Steam for Linux that today, there are a lot of games that I can play. Before November of 2012, I've been dual-booting between Windows and Linux. I got tired of dual-booting ever since.

I did not want to go off-topic, but that is my side of story.

[CrocoDuck]

As far as the sound quality compared to older generations, I can't really say. I will say though that it does sound good so far. Interestingly, (..if my memory serves me correctly), it doesn't seem to sound any better than my old Emu 1616m (which also went up to 192k).

All soundcards are built to have an uncoloring flat frequency response and low nonlinear distortion. This is pretty easy to achieve nowadays unless one does not use the cheapest components in the market and assemble them in the cheapest way. I don't think there are soundcards so utterly unflat and nonlinear you can actually hear the difference from another soundcard. Also, you cannot hear the difference between sample rates at whatever rate higher than 44.1 kHz, unless something odd happens. This is a well proven fact nowadays. In few words: as long as the transfer characteristic is concerned all the soundcards are pretty much the same. What makes the difference is more or less the number of inputs, connectivity, the controls, the quality of converters, clocks and port interfaces. Which means that phrases like "award winning preamplifier" or similar are just plain and simple marketing.

I attached a plot of the frequency response of my Scarlett 2i4 as measured by me with Pseudo-random noise and cross correlation methods, with the card operating at 96 kHz. I am gonna make a full review at some point and I will share the Julia code I used to produce the measurement. You can see what I mean. The preamplifier is pretty much just a flat filter. However, the differences with the peak magnitude response at 20 Hz and 20 kHz are 5 dB. Not really 0.1 dB... Unless they meant something else for 0.1 dB, which they should specify. Phase response should ideally be a straight line to avoid phase distortion. It is not the case with the Scarlett, which seems to have more like two different linear trends jointed at something like 100 Hz. I don't know why they did it (or not even if it is desired). At this point I am also still testing my little program, so pick this with a grain of salt. But yeah, expect these sort of things for every soundcard.

Scarlett2i4.png (34.55 KiB) Viewed 2002 times

Oh - there's one feature on the 2i4 that I think is cool: The Direct Monitor isn't a switch, it's a knob. So you can monitor the signal anywhere in between your direct input and what's coming back from the computer. This could come in handy if you're in a situation where the latency is getting really bad, you could dial-in some of your direct signal. Or even just for troubleshooting.

I am not happy with that knob. The fact is that the signal leaks. If you turn it all the way to input you can still hear part of the playback. It is faint, but heavily distorted and noisy. Not faint enough to be completely masked and it is pretty annoying. I think they just made the mixing knob in a very cheap way. It is actually better to not turn it all the way: just before the end the residual signal is fainter and less distorted...

[CrocoDuck]

However, the differences with the peak magnitude response at 20 Hz and 20 kHz are 5 dB. Not really 0.1 dB... Unless they meant something else for 0.1 dB, which they should specify.

Actually, I think they meant that 20 Hz and 20 kHz are the -3 dB points (cutoff frequencies). In this case the plot should make sense with respect what they declare. What they mean with +/- 0.1 dB is still unknown...

[sadko4u]

Is it possible to look at the full specifications published on the official site? Currently I see only promotion page.

[CrocoDuck]

Is it possible to look at the full specifications published on the official site? Currently I see only promotion page.

The closest thing to full specifications I found for this series is at the bottom of this page and in the manual.

[sadko4u]

The closest thing to full specifications I found for this series is at the bottom of this page and in the manual.

Hmmm. Possible it is on the UK version of the site. I couldn't find it on the General version of the site.

I think this:

20 Hz - 20 kHz ± 0.1dB

Means that the frequency response in this frequency range is floating maximum up to ± 0.1dB.

[CrocoDuck]

I think this:

20 Hz - 20 kHz ± 0.1dB

Means that the frequency response in this frequency range is floating maximum up to ± 0.1dB.

Yep, but it is not absolutely clear. My old generation Scarlett 2i4 had the same specifications. You can see the plot in the previous post. So, 20 Hz and 20 kHz could be the - 3 dB point. It is ok, frequency responses are usually given using - 3 dB points as references. You can see that from 20 Hz the response grows, then hits its maximum, and then decreases again. There are many ways this could have been achieved. The slopes could have been steeper, for example. So +/- 0.1 with respect what? With respect what order filter having the said bandwidth? There are many ways to realize a filter with the same bandwidth that actually has far greater local deviations from the curve above than 0.1 dB, especially towards the -3 dB points. Maybe it is a statistical thing. It could be that 0.1 dB is the standard deviation of the frequency responses of 100 Scarlett 2i4 samples...

[spid4]

Hi All,

I just posted in the hardware list about this as well. I've just got a 2nd gen Scarlett 2i4 and it is working very well under Qiana Studio as described here: viewtopic.php?f=4&t=15747. It gives me occasional xruns at 96000 Hz, 128 sample buffer, and 3 periods per buffer. Cadence says its latency is 1.3 ms.

So that's good news.

[CrocoDuck]

Cool! If you don't mind, it would be nice to know the roundtrip latency as measured by jack_iodelay, as the one reported by Cadence is the math derived latency and applies only at the software level.

[sadko4u]

And what about higher models? Did anyone probe them?

[spid4]

Here's the output from jack_iodelay.

Firstly, at 48000Hz, 64 buffer, 3 periods/buffer:

Code: Select all

451.394 frames      9.404 ms total roundtrip latency
	extra loopback latency: 195 frames
	use 97 for the backend arguments -I and -O

At 96000Hz, 128 buffer, 3 periods/buffer:

Code: Select all

853.929 frames      8.895 ms total roundtrip latency
	extra loopback latency: 341 frames
	use 170 for the backend arguments -I and -O

Those are the two configurations suggested here: http://wiki.linuxaudio.org/hw/focusrite_scarlett

I also tried as 192000Hz, 1024 buffer, 3 periods/buffer. It works but there do seem to be more xruns:

Code: Select all

6728.191 frames     35.043 ms total roundtrip latency
	extra loopback latency: 2632 frames
	use 1316 for the backend arguments -I and -O

Let me know if there are any other tests that would help. I was using it as a guitar stomp box yesterday and the 10ms latency was not a problem at all.

[CrocoDuck]

Interesting results, thanks. It seems that the soundcard is indeed compliant with Focusrite claims, as the roundtrip latency associated with the hardware itself (not coming from JACK, ALSA, software stuff) is pretty much 4 ms at 48 kHz and 96 kHz. It raise up to 14 ms at 192 kHz tho.

Here how it looks like on my older Scarlett:

48000Hz, 64 buffer, 3 periods/buffer:

Code: Select all

   616.685 frames     12.848 ms total roundtrip latency
	extra loopback latency: 360 frames
	use 180 for the backend arguments -I and -O

96000Hz, 128 buffer, 3 periods/buffer:

Code: Select all

  1172.734 frames     12.216 ms total roundtrip latency
	extra loopback latency: 660 frames
	use 330 for the backend arguments -I and -O

My old one doesn't go up to 192 kHz...

We see that the roundtrip latency of the old series is twice (7 - 8 ms). Good to know they improved for real.

Let me know if there are any other tests that would help.

Hopefully I will have my Julia testing code ready for publication soon(ish). At the point you could run it to have the frequency response of the soundcard, so that we can compare with the old series.

[gimmeapill]

That is very interesting, thanks both for testing

So if interpret the results:
The line that tells about the actual HW latency of the card is:

Code: Select all

extra loopback latency: 660 frames

which went down to 341 frames with the 2nd gen = roughly 12 -> 8 ms round trip latency.

As I understand it, it should then be possible to run the 2nd gen at 96000Hz, 128 buffer, 3 periods/buffer, and get roughly the same round trip latency as the first gen running at 96000Hz, 64 buffer, 3 periods/buffer = 8 something ms.

getting similar latency with 128 buffer size vs 64 - This could help significantly with xruns...

Did I get it right?

I'll run again the test this evening with my 2i2 first gen at my preferred setting 96000Hz, 64 buffer, 3 periods/buffer see what it says.
Could you guys give it a try as well?