Audio Interface - Low Latency Performance Data Base :

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

Moderator: Moderators

Post Reply

Re: Audio Interface - Low Latency Performance Data Base :

Post by Goddard »

Pete,

Re RMAA, just have a browse here and here. And see this too.

Vin,

Agree entirely. RTLs alone are just for scoreboarding about being the fastest buffer stuffer.

Turns out my earlier conjecture that Zoom's USB3 AIs were using CEntrance drivers was wrong, they are using drivers developed by a Japanese system house called Interface. For some interesting info (if you can decipher the "Japlish"), run this URL through Google's translater:

http://av.watch.impress.co.jp/docs/series/dal/20150727_713546.html

Btw, that Japanese review did remark about getting a lower RTL under USB2 compared with USB3 (just run that URL though the translater too).

The reason I'm surprised about that 24-sample setting option being available (even if it proves out to just be unusable window dressing) is because it's smaller than the minimum buffer size specified by the ASIO driver as listed in that Russian review (presumably as queried by RMAA's recently-added ASIO facility).

Hey, have you seen MS' info about low latency audio under W10?

Now if only MS would finally offer in-built support for USB Audio Class 2... :frown:
User avatar
Goddard
Frequent Poster
Posts: 993 Joined: Wed Apr 04, 2012 12:00 am

Re: Audio Interface - Low Latency Performance Data Base :

Post by Pete Kaine »

Thanks, NWAVguy, that's a name I've not seen crop up in a while.
User avatar
Pete Kaine
Frequent Poster (Level2)
Posts: 3165 Joined: Thu Jul 10, 2003 12:00 am Location: Manchester
Kit to fuel your G.A.S - https://www.scan.co.uk/shop/pro-audio

Re: Audio Interface - Low Latency Performance Data Base :

Post by TAFKAT »

Goddard wrote:

Turns out my earlier conjecture that Zoom's USB3 AIs were using CEntrance drivers was wrong, they are using drivers developed by a Japanese system house called Interface. For some interesting info (if you can decipher the "Japlish"), run this URL through Google's translater:


Man that was a tough slog, but from what I can make out its more than just the driver, they developed their own proprietary audio specific USB3 controller.

Why aren't the other audio manyfacturers all over this like a rash , or maybe some are .. :-)



Yes I am across it and have had some lengthy chats about it to my MS liaison , short answer, it won't be replacing ASIO any time soon if ever IMO. Great for replacing ASIO4All in mobile environments sans interface , but don't really see anything past that in the near future.

V.
TAFKAT
Regular
Posts: 244 Joined: Wed Jan 08, 2003 12:00 am Location: Australia
AAVIM Technology
DAWbench.com

Re: Audio Interface - Low Latency Performance Data Base :

Post by TAFKAT »

The Zooms have sparked my interest, I am in the process of securing one for testing.

Its great to see the Japanese firms assigning time and energy to the area of dedicated audio controllers/drivers for better latency performance, instead of just falling into line with the off the shelf solutions many settle for. Yamaha also developed in-house for the Steinberg UR line , and despite it not being a class leader, it is better than the generics.

Lets see where the dust settles on this one.. :-)

V.
TAFKAT
Regular
Posts: 244 Joined: Wed Jan 08, 2003 12:00 am Location: Australia
AAVIM Technology
DAWbench.com

Re: Audio Interface - Low Latency Performance Data Base :

Post by TAFKAT »

Hey All,

Not much happening in regards to testing over the last 5 months or so , I will do a new round of testing on the new batch of interfaces released recently when I get them across the bench , but for now just a final follow up post on the Presonus 192. Personally I had hoped to have closed the book on this one and moved on after the initial saga , but the 1.5 and 1.5.1 drivers have delivered a few more curves that need to be highlighted for those reading in and interested in the unit.

Image

The above charts list all the numbers from the initial 1.4 driver to 1.5 and 1.5.1.

So what exactly are we looking at ?

First graph is the reported and delivered I/O and RTL in the Minimal Safety Buffer mode which was manually available on 1.4.*. The second graph ( Consolidated ) is the recommended matched pairings for Safety / Playback buffer , which are now hard set on the 1.5 driver. Third graph is the release 1.5 driver , fourth the 1.5.1 driver/firmware with the widely reported lower latency.

First thing to note, using the recommended matched pairings measurably increases I/O and RTL on anything above 064 where Minimal Safety is hard set , so 1.5 did not improve I/O and RTL on any of the other settings, it made them measurably higher. The new 1.5.1 driver still has the locked pairings and subsequently has hight I/O and RTL than the initial 1.4 driver on all settings above 064. The matched pairings are supposed to improve inoculation against drop outs due to the overall higher playback latency, well that's the theory.

In practice the higher USB kernel block safety buffer is instead collapsing the driver performance once the processing overhead is scaled up on larger sessions. This goes against the theory of higher playback latency lessoning the chance of drop outs, except there is another variable introduced by using very large USB block buffers , which was not tested and accounted for, IMO. In all my testing on the various interfaces using the same OEM driver, the driver is best at the lowest safety buffer setting, across all playback buffers , not under the matched pairings.

Why haven't I reported this earlier seeing the 1.5 driver has been out since end of February ? Simply because the performance went backwards and I was way over all of the melodrama over this release and want to remove myself from the ongoing saga over the relevance or non relevance of the driver performance.I held off hoping that the 1.5.1 driver/firmware release would address the issues and return the unlocked pairings to at least have the option to run at some improved latency settings.

To say I was disappointed that after 6 months of development and 2 major revisions, that the driver performance went backwards , is an understatement.

Those that require efficient low latency performance and can understand the data provided will have their answer , cyclic debates re personal preference and relevance have already been covered.

As I clearly stated in an earlier post , latency performance is more than delivered playback latency and RTL, its how efficient the drivers are at the respective working latencies , the charts above are a clear indicator of where the current Presonus driver is heading in that regard.

V.C
TAFKAT
Regular
Posts: 244 Joined: Wed Jan 08, 2003 12:00 am Location: Australia
AAVIM Technology
DAWbench.com

Re: Audio Interface - Low Latency Performance Data Base :

Post by Martin Walker »

Thanks for the update Vin - how sad that performance is going backwards in this day and age :frown:

Martin
User avatar
Martin Walker
Moderator
Posts: 18310 Joined: Wed Jan 13, 2010 8:44 am Location: Cornwall, UK

Re: Audio Interface - Low Latency Performance Data Base :

Post by TAFKAT »

Martin Walker wrote:Thanks for the update Vin - how sad that performance is going backwards in this day and age :frown:


Very disappointing and increasingly frustrating , hopefully this is not going to become a common theme with the manufacturers using these OEM driver. All I can do is point it out if they do.

The information re the locked pairings hobbling performance at the expense of higher playback latency was something that I reported to Presonus 6 months ago , but was obviously rejected and dismissed.

You can lead a horse to water.

:beamup:

V.
TAFKAT
Regular
Posts: 244 Joined: Wed Jan 08, 2003 12:00 am Location: Australia
AAVIM Technology
DAWbench.com

Re: Audio Interface - Low Latency Performance Data Base :

Post by TAFKAT »

TAFKAT wrote:
The information re the locked pairings * hobbling performance at the expense of higher playback *


Doh, that was worded badly, what I meant was * hobbling performance in addition to higher playback latency *

V.
TAFKAT
Regular
Posts: 244 Joined: Wed Jan 08, 2003 12:00 am Location: Australia
AAVIM Technology
DAWbench.com

Re: Audio Interface - Low Latency Performance Data Base :

Post by ef37a »

Well I for one am mad as hell.

I have never had a PC with Fussywire and thought USB 3.0 was going to be a viable alternative to TB? Now it seems that one of the first USB 3.0 audio products from a major player is a total pig's ear as far as latency goes. Grrrr!

I am still waiting for some enterprising mnfctr to produce "budget" 3.0 AIs with better bus powered performance (for USB 2.0 the features could be curtailed or a PSU used)

Dave. (Still ticking dem damn boxes!)
ef37a
Jedi Poster
Posts: 13262 Joined: Mon May 29, 2006 12:00 am Location: northampton uk

Re: Audio Interface - Low Latency Performance Data Base :

Post by awjoe »

ef37a wrote:I have never had a PC with Fussywire and thought USB 3.0 was going to be a viable alternative to TB? Now it seems that one of the first USB 3.0 audio products from a major player is a total pig's ear as far as latency goes. Grrrr!



I've just spent all morning in this thread and its links. Must get a life...

I had high hopes for USB 3, too. But it looks like my best bet is to hang onto my Presonus Firestudio Mobile for a while yet. It just about ticks all the boxes for now.
User avatar
awjoe
Frequent Poster (Level2)
Posts: 3349 Joined: Tue Mar 08, 2011 12:00 am
The worst is elevator versions of good songs. All the life sucked out of them in a most amusing and yet irritating way.

Re: Audio Interface - Low Latency Performance Data Base :

Post by TAFKAT »

LLP Database Update: August 2017:

Some additions, updates and amendments.

Image
Image

Added:

RME Fireface UFX+ : This is RME's flagship in the Fireface line, offering Thunderbolt 2 and USB3, FW has been retired. As has come to be expected from RME, low latency performance and efficiency is class leading across the 2 available protocols. There has been a fair amount of discussion re Thunderbolt potentially having PCIe level performance and RME deliver performance that is very close to par to the reference HDSPe PCIE solutions, at an even slightly lower overall I/O and RTL. USB3 is equally impressive , shadowing the TB2 results , and also offering a 032 buffer setting over their USB2 offerings. As I had noted in an earlier post , we are essentially splitting hairs over the 3 protocols in regards to delivered performance from the RME units. Very impressive.

Presonus Quantum : This is a unit that has generated a lot of interest at launch with the claims of its ultra low latency , lack of any DSP/hardware based monitoring , with all the dependency being on the drivers efficiency. So have they delivered, well in regards to low latency performance they certainly have come full circle. Performance at the latencies within the testing parameters were well within range of the best available at even lower I/O and RTL latencies , very impressive to say the least, especially after the curves navigated with their previous top line 192 USB3 unit.

The unit does have buffer settings available down to 08 samples , I did test at both 08 and 16 samples , and was able to play the reference session for DSP and add some plugins , but neither were really practical for VI's. Thats not to say you could not experiment with the settings for input monitoring in lower resource sessions to get those super tight RTL's , but IMO they are really only there to compensate the fact that the unit is devoid of any hardware direct monitoring facilities. YMMV.

I/O and RTL listed below for reference.

08 Samples : Reported I/O - In 0.363 / Out 0.544 : Measured RTL - 0.995

16 Samples : Reported I/O - In 0.544 / Out 0.726 : Measured RTL - 1.361

I am very happy to see that someone at Presonus has been listening, taking notes and have delivered!

UAD Apollo Twin USB3 : A few curves navigated installing the software on Windows 7, further curves getting the unit to work on a 3rd party USB3 card on the reference system, but all sorted without too much blood loss.

I tested the units RTL - Round Trip Latency with Input Delay Compensation Off, to minimize input latency , default is Medium. The Input Delay Compensation settings introduce additional latency on the Input to smooth out the DSP processing when used on the input to either monitor or commit/print, so really needs to be switched off unless using UAD on input.

Input Delay Compensation Values below.

Short : + 100 Additional Samples on Input.

Medium : +200 Additional Samples on Input.

Long : + 1000 Additional Samples on Input.

Output latency remains unchanged.

Obviously these substantially increase RTL when monitoring via software if left enabled.

Low latency Performance is pretty much as I expected, below average , ( I had already been warned by some colleagues who will remain nameless ) so for anything above basic tracking it falls over pretty quickly when adding Plugins, Virtual Instruments. I didn't go into testing the DSP side of the unit, as that wasn't my focus.

I know those requiring LLP are not the target market for the unit, and the results will pretty much be dismissed by most who will gravitate to the unit.

All good, horses for courses.

Focusrite Scarlett 6i6 G2 : The Gen 2 units stepped up and delivered significantly improved performance over their previous Gen1 offerings. Again this is a case of the manufacturers listening and delivering.

Performance is quite respectable but I did have to navigate the odd buffer setting labeling which essentially is one step out. I have chosen to post the results with what I termed the corrected values with actual panel settings listed in brackets, as it made more sense to the comparative performance against the other interfaces tested.

I have been in communication with the team at Focusrite re possibly amending the labeling to indicate the correct values, but to be honest I never followed it up past the inital discussions and Focusrite have not followed up or through in the amendments, so just take note of the panel settings and the true delivered I/O and RTL

Audient iD4 : All the Audient interface units use a unified driver package , which one would assume would deliver identical performance across the range, but one would be wrong unfortunately. The iD4 under testing experienced very different levels of performance using the same driver package when tested against the iD22 which I have here as another reference unit.

All units are using the standard OEM Thesycon driver package , however on the iD4 , Minimum safety buffer resulted in a total collapse of the driver regards playback of even the base templates of the test sessions , so no results were achieved. I raised the safety buffer up one step which also disabled the lowest available buffer of 064 , raised I/O and RTL and tested only at 128 on. Same driver package was used successfully on the iD22 at Minimum safety and I have updated the results.

Updated results:

Audient iD22 : As noted in the notes for the iD4, I ran the sessions on the new updated unified driver , results were actually a little down from the previous driver. I was not overly surprised as Thesycons latest offerings have been going in the wrong direction regards performance for a while.

Presonus 192 : This is a unit I have spent way too much time and energy on as anyone who has been following the soap opera surrounding the driver performance would be aware. You will note that the driver version I have stopped testing at is quite a few revisions back and it is the first one that introduced the reported lower latency from the DSP, but also threw out the baby with the bath water by locking the safety and playback buffers into set pairs , which raised I/O and RTL , as well as neutering delivered performance on both the DSP and VI test sessions.

Yes, the performance anomalies of locking the buffer settings to matched pairs was forwarded to the Presonus team very early on , was noted and obviously dismissed or not regarded as an issue. The locked pairings have continued right up to the current version of UC.

The performance will be consistent across all Presonus USB2 devices using UC.

That's it for now, I'll stick my head up again after I have some tested some new hardware, or if anything significantly changes with any previously tested.

V.C
TAFKAT
Regular
Posts: 244 Joined: Wed Jan 08, 2003 12:00 am Location: Australia
AAVIM Technology
DAWbench.com

Re: Audio Interface - Low Latency Performance Data Base :

Post by Martin Walker »

You really are a glutton for punishment Vin! 8-)

Two interface manufacturers are currently cheering at your results, while the rest work out how they can take out a contract on you without getting rumbled :beamup:

Thanks for all your efforts on our collective behalf though ;)

Martin
User avatar
Martin Walker
Moderator
Posts: 18310 Joined: Wed Jan 13, 2010 8:44 am Location: Cornwall, UK
Post Reply