You are here

Audio Interface - Low Latency Performance Data Base :

For anything relating to music-making on Windows computers, with lots of FAQs. Moderated by Martin Walker.

Re: Audio Interface - Low Latency Performance Data Base :

Postby TAFKAT » Mon Mar 25, 2013 9:45 pm

Swinging this back..,

I will be in front of an 828 Mk3 Hybrid shortly, so I can finally get some MOTU USB numbers into the Database.

V.C
TAFKAT
Regular
Posts: 244
Joined: Wed Jan 08, 2003 1:00 am
Location: Australia
AAVIM Technology
DAWbench.com

Re: Audio Interface - Low Latency Performance Data Base :

Postby Martin Walker » Tue Mar 26, 2013 7:05 pm

Well done Vin - your plan to take over the world is getting ever closer :D


Martin
User avatar
Martin Walker
Moderator
Posts: 15597
Joined: Wed Jan 13, 2010 9:44 am
Location: Cornwall, UK

Re: Audio Interface - Low Latency Performance Data Base :

Postby TAFKAT » Tue Mar 26, 2013 11:27 pm

Is it that obvious Martin... LOL

:bouncy:

V.
TAFKAT
Regular
Posts: 244
Joined: Wed Jan 08, 2003 1:00 am
Location: Australia
AAVIM Technology
DAWbench.com

Re: Audio Interface - Low Latency Performance Data Base :

Postby Hugh Robjohns » Wed Mar 27, 2013 11:38 am

TAFKAT wrote:not one word on delivered RTL's, not one word on the actual performance at respective latencies , but plenty of chatter about the supposed super low latency of the Dante protocol.


Sorry to be late in on this one -- didn't realise you were discussing the Rednet here and I'm just responding to a link from another thread...

As most of you know, the Dante system has been around for several years now and has a very solid track record in live sound systems where latency is just as critical as in music studio applications. It's a system developed specifically for high-end, high channel-count professional applications, and Focusrite are marketing their implementation of it at high-end professional studios. Poor latency (let alone poor perforamnce) just isn't acceptable in these markets. I know there have been issues at times with some budget Focusrite interfaces, but the Rednet stuff is a collaborative development based on proven existing technology and checked against the implementations of other manufacturers.

By design, the Dante IP network latency is tiny -- absolute worst case is about 0.5ms, and more typically we're talking 0.2ms or less. Converter latencies are the same as for any other interface, of course, so the defining factor really comes down to the computer set up and the specific host interface in use.

The Dante Virtual Soundcard solution is the freebie starting option and while it's okay and usable, if low latency is a really important factor then you wouldn't start there. The dedicated Focusrite PCIe card -- because of the fact that it sits directly on the PCIe bus -- is extremely fast and on a par with the best internal ASIO cards. Soemthing like the RME MADI card would be a fairer comparison, but I use an RME AIO on my own primary audio computer and the Rednet system seemed to perform at a very similar latency level. I measured a real-world round trip latency analogue in to analogue out of about 3ms with the system set up straight out of the box at 44.1kHz sample rates. At higher sample rates the latency shortened as you would expect.

I didn't delve into that aspect further because I saw no need. It clearly worked extremely well, and latency wasn't an issue at all as far as I was concerned. I also ran a fairly large (by my standards) Reaper project with 46 tracks and a fair fistfull of plugins on the system, contriving to send lots of tracks asdn mixes out via the Dante interfaces and back again and had no problems whatsoever.

It may have been possible to improve further on the latencies by tweaking the operating system etc -- and Focusrite have some advice about that on their website -- but I didn't bother as there was no need. The host PC was supplied by Focusrite and was obviously a high spec model that had been set up for Rednet... but that's fair enough as far as I'm concerned.

I don't think Rednet is really designed for small home project studios, and it seems to me that it only really becomes cost effective for seriously large system with high channel counts and the need to interface I/O boxes in far flung places -- such as in a larger multi-studio complex.

Hope that helps

H
User avatar
Hugh Robjohns
Moderator
Posts: 27005
Joined: Fri Jul 25, 2003 12:00 am
Location: Worcestershire, UK
Technical Editor, Sound On Sound

Re: Audio Interface - Low Latency Performance Data Base :

Postby Goddard » Wed Mar 27, 2013 12:36 pm

Hugh, thanks for your input. Hadn't realized you'd missed this party, had assumed we had you to thank for Neil's joining in earlier.

The burning question for some of us is how that 3ms latency figure holds up (scales) under practical workload scenarios such as (those simulated by) Vin's DAWbench LLP test suite.

I think everyone pretty well grasps what the intended target market is, but Focusrite have posted performance data for a number of tested systems employing the PCIe card in a tracking setup and giving track count and sample rate at a specified 3ms latency, and some of us are very curious how that performance holds up in "practical" scenarios beyond simply tracking I/O in Cubase (so your input here on your usage scenario during your testing is appreciated).

And even in the tracking scenario reported by Focusrite, there is quite a performance disparity apparent in the levels attained at that 3ms latency by the tested Mac Pro Xeon 6- and 8-core systems, the two tested Xeon E-3v2 4-core systems and the tested Rain overclocked i7-3770k 4-core system, which leaves some of us (well, me) wondering.

I suppose that, thanks to Martin giving DAWbench LLP info in his reviews and the value thereof to those of us concerned with practical performance, we've now grown to expect such info from SOS.
User avatar
Goddard
Frequent Poster
Posts: 999
Joined: Wed Apr 04, 2012 12:00 am

Re: Audio Interface - Low Latency Performance Data Base :

Postby Hugh Robjohns » Wed Mar 27, 2013 12:56 pm

Goddard wrote:The burning question for some of us is how that 3ms latency figure holds up (scales) under practical workload scenarios such as (those simulated by) Vin's DAWbench LLP test suite.

Yes, I appreciate why people are intested in this.... but I can't answer it because I didn't run a DAWbench test. All I can say is that I tried to run as big a practical real-world test on it as I could, and it performed flawlessly. I had four hardware interfaces hooked up to the PCIe card, and I had audio flying around all the outputs and inputs, and it all just worked.

I suppose that, thanks to Martin giving DAWbench LLP info in his reviews and the value thereof to those of us concerned with practical performance, we've now grown to expect such info from SOS.

Different reviewers have different skillsets. As I'm sure you'll realise, computer interfaces aren't an area that I usually get involved in. I was asked to review the Rednet system because it isn't directly comparable to normal USB/FW computer interfaces, is intended for the high-end pro-market, and the review emphasis was more about what the new technology is and what it offers. I'm sure we're going to see further reviews of this system as time goes on...

H
User avatar
Hugh Robjohns
Moderator
Posts: 27005
Joined: Fri Jul 25, 2003 12:00 am
Location: Worcestershire, UK
Technical Editor, Sound On Sound

Re: Audio Interface - Low Latency Performance Data Base :

Postby Patrice Brousseau » Wed Mar 27, 2013 1:55 pm

Tombot wrote:
Goddard wrote:Hi Pete, thanks for those numbers.

Was that using the version 1.50 driver under Win7, or the current v1.51 (under Win8?)?

I'm curious, because I've seen a report of very low latency at 32 samples, although afaik Roland only ever claimed 48 samples (~1ms) as smallest "VS Streaming" buffer size, at least prior to releasing the v1.50 driver.

Best results came without VS Streaming ticked. it just seemed to increase the rtl times tbh.

Hi all,

Measured RTL with Oblique RTL utility: 7.659ms at 64 samples. This is with 1.5 drivers, option "Reduce CPU load" ticked (without this option, the figures goes around 9.5ms).

On a MacBookPro, Win7 SP1 x64 in Bootcamp.

BTW, the interface could go as low as 32 samples but I don't know the real efficiency with this low latency...
Patrice Brousseau
Poster
Posts: 20
Joined: Tue Sep 14, 2004 12:00 am
Location: Montréal, Canada
Patrice Brousseau

Re: Audio Interface - Low Latency Performance Data Base :

Postby Goddard » Thu Mar 28, 2013 8:29 am

Hugh Robjohns wrote:
Goddard wrote:I suppose that, thanks to Martin giving DAWbench LLP info in his reviews and the value thereof to those of us concerned with practical performance, we've now grown to expect such info from SOS.


Different reviewers have different skillsets. As I'm sure you'll realise, computer interfaces aren't an area that I usually get involved in.


Just put it all down to Martin setting the bar so high, Hugh! :D

And perhaps also, to Vin's "no eggshells spared" criticism when expressed latency figures don't reflect LLP evaluation. :bouncy:

If Neil is still around, he might be interested to know that my recent decision to purchase a Saffire Pro24 DSP was based on info provided by both you (your SOS evaluation of its VRM capabilities in which, besides criticism of Focusrite's lack of high end speaker/environment model inclusion you'd also opined as to its suitability for headphone mixing) as well as Vin's confirmation of the improved LLP performance of the Pro24 with the updated Dice FW drivers, all of which info was spot on and eminently helpful in my interface selection process, so I do look to you both for the "good" info before opening the purse.

To put the above tale into perspective, another. My acquisition of a TC Konnekt 24D a good while back had been based in part on a review in SOS also (despite the problems reported therein by the reviewer, and largely because of the promise of the nice in-built TC fx), but I had nothing but problems with drivers and software and compatibility, and despite the fx it never quite lived up to its promise (perhaps why it was soon after withdrawn from the market). Recently, as my requirements expanded, I'd looked in the direction of TC's newer, top o' the line Konnekt 48 rack unit, but then found that users have been reporting problems with, (surprise!) the drivers and software. So, shall wait to see whether it passes muster with both SOS and DAWbench before casting a glance in TC's direction again. ;)
User avatar
Goddard
Frequent Poster
Posts: 999
Joined: Wed Apr 04, 2012 12:00 am

Re: Audio Interface - Low Latency Performance Data Base :

Postby TAFKAT » Sun Apr 07, 2013 12:12 am

Quick heads up,

I have completed the testing on the MOTU 828Mk3 USB ( couldn't redo the FW because MOTU in all their wisdom only supply a FW800 cable- WTF , obvious where their target marketing focus is )and have updated and amended the charts to also include the UCX and also the new V3 drivers for Presonus/Focusrite ( see G I was paying attention ).

I am also wanting comment in detail on some points raised in Hugh's response re the Focusrite Rednet, but I'll hold off until I put the final touches on the Database.

Peace

V.C
TAFKAT
Regular
Posts: 244
Joined: Wed Jan 08, 2003 1:00 am
Location: Australia
AAVIM Technology
DAWbench.com

Re: Audio Interface - Low Latency Performance Data Base :

Postby Goddard » Mon Apr 08, 2013 3:11 am

:D
User avatar
Goddard
Frequent Poster
Posts: 999
Joined: Wed Apr 04, 2012 12:00 am

Re: Audio Interface - Low Latency Performance Data Base :

Postby Patrice Brousseau » Fri Apr 12, 2013 1:26 pm

Pete Kaine wrote:Right then, I'll throw a quick comment into this thread, althrough I'll need a chat with Vin before we get them up properly.

The Octo came out at a score of 5.3.

RTL:
7.6 @ 32
12.09 @ 64
20.75 @ 256

So perfectly usable at 32 althrough real time monitoring might become a little questionable at 64 and above.

With RTL Oblique util:

7.659ms at 64 [email]samples@44.1kHz.[/email] This is with 1.5 drivers, option "Reduce CPU load" ticked (without this option, the figures goes around 9.5ms).

48 buffers: 6.4ms

On a MacBookPro, Win7 SP1 x64 in Bootcamp.

I repeat, if you don't tick "Reduce CPU load", the figures goes higher. Maybe also that the 1.50 driver is better optimized in this regard compared to the new 1.51.

Just to be sure, I've also ran a loopback test inside Reaper and the RTL figures are the same.

**As a sidenote, AU Lab (an utility, AU and VST Host under OSX) reports exactly the figures, confirmed by loopback test.

"AU Lab : hardware input : 136 samples safety offset : 22 samples
hardware output : 4 samples safety offset : 22 samples"
Obviously, you add 64 in and 64 out and by calculating number of samples: 312 samples or 7.1ms. RTL figure is a bit lower in CoreAudio but probably that Reaper won't be unable to sustain the same loads as under Win7 x64.

Patrice

BTW, the interface could go as low as 32 samples but I don't know the real efficiency with this low latency...
Patrice Brousseau
Poster
Posts: 20
Joined: Tue Sep 14, 2004 12:00 am
Location: Montréal, Canada
Patrice Brousseau

Re: Audio Interface - Low Latency Performance Data Base :

Postby Patrice Brousseau » Fri Apr 12, 2013 1:38 pm

BTW,

Echo Mia, drivers 6.08 wrapped under ASIO4ALL 2.7:

On XP 32 bits: 64 [email]buffers@44.1[/email] kHz, 6.1 ms RTL, not bad.

Edirol FA-66 also wrapped under ASIO4ALL: 64 [email]buffers@44.1[/email] kHz, 8.8ms RTL. Usable but not impressive.

Now, for those asking why using ASIO4ALL? With native Echo Asio drivers, RTL is 5.7ms but the setting is almost unusable as it "crackle-pops" easily.

The Edirol: lowest Asio buffer, 88 giving 10.something RTL... Forget software monitoring!

And why XP32 on my old (Q6600, 4 Gb's RAM) desktop? It performs flawlessly as low as 64 buffers without worrying about instabilities or drop-out. Just make music...

I would like for sure to add a streamlined Win7 x64 DAW boot but so far, the Edirol and Echo results under my standard (Web, Office) boot are less than stellar (crackles, midi playing sluggish with just 128 buffers. This was with all unnecessary services and processes disabled, classic theme, latest drivers...

Patrice
Patrice Brousseau
Poster
Posts: 20
Joined: Tue Sep 14, 2004 12:00 am
Location: Montréal, Canada
Patrice Brousseau

Re: Audio Interface - Low Latency Performance Data Base :

Postby TAFKAT » Fri May 03, 2013 11:43 pm

Hey,

Its been a while but I have updated the database , added a few , removed a few and will give a quick review of the new entries.

Image
Image

Additions :

ESI Julia XTe : What a pleasant surprise packet this little card turned out to be after the mixed result of the earlier MAYA44e card. Lower RTL than the RME base reference for the working latencies , respectable performance across the 3 benchmarks placed it well in the upper registers of the LLP table. Its still not without a few quirks, one being the reported latency is nominal for In/Out, so we really aren't entirely sure what the real deal is there. the latency variance is between reported and calculated is 54-69 samples so that could be combination of the AD/DA and some added buffers, but no way to really check 100%.. :-(

Presonus Firestudio Mobile : Another one that was a very pleasant and welcome surprise with the new Version 4.0 driver from TC Applied. So someone was listening, in short a great performer now only bettered by the RME's. Very solid performance across all benchmarks and working latencies. The version 4 driver is a unified driver across all the range, so the improved performance is valid for the larger Firestudio Project and the Digital Mixers as well.

Focusrite Saffire Pro 24 : Ditto the above , using the same TC applied supplied Version 4.0 driver as Presonus. Performance across the benchmarks was identical to the presonus, the RTL was a little better so it achieved a higher RTL. Again this performance is valid across all the Focusrite FW range.

MOTU 828 Mk3 Hybrid : Where do I start , the unit is a FW400 / USB2 Hybrid interface that MOTU have not had the wisdom to even supply a FW800 to FW400 cable with , very obvious that the target market is not Windows , but lets not get to caught up with that. The unit is based on a new ArchWave controller and uses a unified driver install for both FW/USB2 which reported identical In/Out values for both FW and USB2 to Cubase which immediately threw up a bit of a flag. On using the RTL Utility the values were measurably different so already we have some grey area to navigate. The RTL Utility was also reporting an error in the calculation because the value was over 50% different to what was being reported by the driver , enter hidden buffers , not to mention the double buffering on playback to round off the less then stellar RTL values.

All that aside, how did it perform. FW performed almost equivalent to the previous non Hybrid , but not as good in RTL or benchmark results , the USB when I could arm-wrestle some consistency , performed measurably worse than the FW. Overall, pretty average.

Mackie Onyx 1640i : Heres one I had been sitting on for a while as I didn't have the measured RTL values and the reported nominal values were, hmmm, very optimistic. I finally had the opportunity to get in front of one again to do the RTL calculations. Wow, in short the settings values have absolutely no correlation to what is delivered and even more concerning was I could not get a consistent result from the RTL utility, with the value swing around up to 80 samples at a time. Check out the cart for the actual delivered RTL , I have no idea how the DAW's could possible compensate for that amount of variance , absolutely unworkable IMO.

Removed :

Presonus Studio Live/ Focusrite Saffire 56 :
With the new version 4 unified driver I will update the individual units when I get the opportunity , but for now the new enties will give a good indication of where they sit in the LLP table.

MOTU 828 Mk3 FW : No longer current, tables updated to the new Mk3 Hybrid.

Tested but not included :

Focusrite Scarlett 18i6 : I tried very hard to get a result for this interface but it threw up a few challenges. First off the buffers are set to ms so I didn't have the required range of buffer settings, they also have absolutely no correlation to what is delivered. The reported values show double buffering on input, triple on playback and to top it off, the calculated RTL was lower than than that reported by the driver. If allowed a bit of grey to allow the 2 lower buffer settings of 1ms/2ms to replace 032/064 I came up with an overall LLP in the low 4's , RTL's ranged from 6.858 at the supposed 1ms value all the way to 61.997 at the supposed 12ms. That should give everyone enough of an idea of where the interface sits in the greater scheme.

I said earlier that I was going to comment in detail on Hughs responses re RedNet but on further consideration I am going to reserve comment until I have the opportunity to test it myself. There is just so much grey in the commentary regarding what the actual latency values are being reported ( RTL or Direct Monitoring ) that it really needs to be 100% clarified, which cannot be accomplished without being in front of the units again. There is also the question of how it actually performs at the respective working latencies as well as the so called intended market , but I'll cover that at the appropriate time if/when I can get a chance to test the units.

Peace

Vin Curigliano
TAFKAT
Regular
Posts: 244
Joined: Wed Jan 08, 2003 1:00 am
Location: Australia
AAVIM Technology
DAWbench.com

Re: Audio Interface - Low Latency Performance Data Base :

Postby Goddard » Sat May 04, 2013 2:26 am

Good on you, Vin! Thanks for the update!

Wrt the ESI Julia XTe, wondering if the contemporary Maya44 XTe (updated version of Maya44e) performs similarly?

As you've not included Scan's Roland Octa test numbers (due to test system differences?), let's hope that Martin or perhaps Craig A. ;-) might persuade Roland to send some interfaces your way.

And while you're waiting on Rednet, keep an eye out for AVB-capable interfaces to materialize (Prism Lyra2?), as Intel are already offering an AVB-capable NIC.
User avatar
Goddard
Frequent Poster
Posts: 999
Joined: Wed Apr 04, 2012 12:00 am

Re: Audio Interface - Low Latency Performance Data Base :

Postby TAFKAT » Sun May 05, 2013 12:18 am

Hey G,

Not sure about the ESI 44XTe , the driver version is slightly different but it could well be better as its a family of driver versions closer to the Julia XTe.

Re the Roland, I have given up asking for a test unit, Roland are obviously not interested .

Re RedNet and beyond, hmmm, lets see what it actually delivers , I may also have a chance of getting in front of an Antelope Orion 32 with its so called uber proprietary USB2 controller, so some interesting days ahead I suspect. I hope the reports of the OSX performance/stability isn't a reflection of the Windows driver.

V.
TAFKAT
Regular
Posts: 244
Joined: Wed Jan 08, 2003 1:00 am
Location: Australia
AAVIM Technology
DAWbench.com

PreviousNext