Audio Interface - Low Latency Performance Data Base :

[TAFKAT]

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 of performance issues. A lot of the qualms have been due to the low latency performance, or to be exact, lack of , so I thought it would be a good idea to place the spotlight on that specific aspect of the current crop of audio interfaces.

By testing the interfaces head to head and reporting warts and all, I hope to place some of these developers on notice that jamming as many shiny baubles onto an interface , but neglecting the crucial aspect of the actual driver performance , isn't going to swing for all of us.

I have been working on this for about 15 months odd so the preliminary results were actually from earlier in 2010 , but are still valid and are a good starting point as I add further to the list over the coming month/years.

Here are some initial results for both Desktop and Notebook .

Desktop System Detail:
Intel i7 920 Quadcore/ 2.66 GHZ/
Intel X58 / 6 GB DDR3-PC12800.

Notebook System Detail:
Intel i7 620 Dualcore / 2.66 GHZ/
Intel HM55 / 4 GB DDR3-PC10600.

O.S Detail:
Windows 7 x64 Pro





All of the interfaces listed except for the baseline RME HDSPe units are FW and its quite obvious from the results that there are large variables in LLP- Low Latency Performance depending on the efficiency of the respective ASIO drivers / FW controllers being employed on the respective interfaces.

I have posted a summary of the initial testing at the DAWbench site Here

I have a second round of testing on the way which will include some of the newer AVID interfaces at both ends of the spectrum , Mbox Pro 3 and also one that I have had quite a few requests to put the thru its paces, the PTHD Native PCIe card.

Stay Tuned



----------------------------------------------------------------------------------------------------------

Latest Results and Downloads :

RTL Utility by Oblique Audio - used to measure Round Trip Latency : Here

Latest Results : 08 May 2012 - Here

[wakestyle]

good stuff.
maybe you can test others like Stienberg, EMU, Roland. Nice to see the M-Audio drivers performing well.

I take it UFX would be somewhere's between fireface and AIO?

[TAFKAT]

Originally Posted by wakestyle

good stuff.
maybe you can test others like Stienberg, EMU, Roland. Nice to see the M-Audio drivers performing well.

I take it UFX would be somewhere's between fireface and AIO?

Hey W,

I'll test anything I get my hands on and post the results.

RME UFX is not actually between the FF and AIO, more so just a touch behind FF800 on FW and a touch behind that again for USB2. I'll get some detailed UFX FW/USB numbers up ASAP , but I do have some RME Babyface numbers , which should be on par with the USB2 numbers.

Desktop :

DAWbench DSP RXC :

Cubase 5.1.2

Core i7 920 2.66 GHZ:

6GB PC12800 DDR 3

Windows 7 x64 Pro

RME : Babyface ; Driver 0.961

048 : 122 RXC : I/O - 2.336/2.517

064 : 125 RXC : I/O - 2.698/2.880

128 : 141 RXC : I/O - 4.150/4.331

256 : 154 RXC : I/O - 7.052/7.234

---------------------------------------------------

Notebook :

DAWbench DSP RXC :

Cubase 5.1.2

Core i7 620 2.66 GHZ:

4GB PC10600 DDR 3

Windows 7 x64 Pro

RME : Babyface ; Driver 0.961

048 : 31 RXC : I/O - 2.336/2.517

064 : 34 RXC : I/O - 2.698/2.880

128 : 42 RXC : I/O - 4.150/4.331

256 : 48 RXC : I/O - 7.052/7.234
----------------------------------------------------
Notebook numbers for the Babyface were very disappointing comparatively on the original reference i7 notebook which I no longer have , I'll revisit it again with my new Quadcore Sandybridge notebook and see if they fare any better.

[wakestyle]

Originally Posted by TAFKAT

I'll test anything I get my hands on and post the results.

coolio.

[TAFKAT]

O.K, here are the results for the first batch of audio interfaces on the new round of testing.




Explanation of how the LLP- Low Latency Performance Rating is derived.

The results for the DAWbench DSP RXC across the latencies of 032 thru to 256 ( which has been the M.O for the last 5 years ) are added and the total is then % wise gauged against the result for the RME HDSPe AIO baseline card. The same is then calculated for the DAWbench VI CV/ NCV tests for 032-512.

Those 3 % results are then added and divided by 3 to give an average % .

I thought it important for the I/O and RTL figures to be an influencing factor on the rating as some cards have a lot lower overall latency than others, so the average % results is then multiplied by the last % result for the RTL.

How the RTL % is calculated is I combine the total of the RTL's across the specific available buffer settings for the cards ( all cards past the base reference RME were calculated using 064-512 except the M-Audio FastTrack which had a range of 128-512. You will notice 2 values listed for RTL for the RME HDSPe AIO , first being for 032-512, second being 064-512 ) , and then calculate the % variable against the baseline. You will see that some cards are actually better than the RME AIO card in regards to I/O /RTL i.e RME Babyface and the AVID MboxPro 3.

I think that is a fair appraisal using the collated data, and it gives deserved credit and advantage to those cards that do have lower individual In/Out and Round Trip Latencies.

Some cards do not report the correct I/O values to the DAW Hosts , only reporting the nominal value for the respective buffer sizes. In those instances I will make a note and default the RTL rating to .95 which is below the average of the other cards RTL % rating of around .97 , not perfect but a reasonable compromise.

A few side notes - the baseline reference is the RME HDSPe AIO card using the 3.08.5 driver. You will notice that even the new version 3.24 driver is also tested and gauged against that baseline result as the performance has dropped. If a later RME HDSPe driver levels or betters the 3.08.5 driver, then it will become the new baseline, unless of course another card has better overall performance and becomes the new baseline.

The performance of the M-Audio Fast Track Ultra was actually quite good for a USB2 interface at the available latencies, but its rating suffered due to the lowest available buffer being only 128. The results for the RME Babyface was very impressive for a USB 2 interface , edging out the FW - Profire 610 in the rating due to its lower overall I/O and RTL. The Result for the Focusrite unit using the dreaded Dice II FW controller , speaks for itself !

More results and analysis coming in the next few weeks

Stay Tuned.

[TNM]

vin do you believe the esi card is reporting it;s latency correctly and has no hidden buffers?

[TAFKAT]

Hey T,

You beat me to the punch.. :-)

I have since discovered its only reporting the nominal settings for the respective buffer settings unfortunately , I'll amend the tables and LLP Rating accordingly.

I am trying to get in touch with ESI to see if I can get the correct values.

[TAFKAT]

Tables amended, LLP Rating adjusted and comments updated

[Jomox]

Would like to see how the 1616m performs against the rest in this test.

[Itsmorning]

Hey tafkat thanks alot for testing the audio interfaces. Your a true contibuter. What's your opinion of the babyface? Looked pretty good in ur tests. Anything important to take into account before purchasing. Thanks. Cheers.

[TAFKAT]

Originally Posted by Itsmorning

Hey tafkat thanks alot for testing the audio interfaces. Your a true contibuter. What's your opinion of the babyface? Looked pretty good in ur tests. Anything important to take into account before purchasing. Thanks. Cheers.

Hey ,

I am focusing only on the Low Latency Performance / Stability of the interfaces, and in that department the Babyface as a USB2 interface delivered admirably , keeping the best of the FW interfaces in sight and besting some others. Remember this is on a current Intel desktop system, notebooks are a mixed bag as evidenced by numerous recent reports here, so you'll need to weigh that up for your own personal working environment.

RE testing the EMU's, unfortunately not something that I have access to.

[zephonic]

Any chance you can get your hands on an MR816, Vin? It would be interesting to see how it compares to the MBox3Pro.

I presume you used Cubendo for the plug-in counts and ms latency test?

[redddog]

Somebody get this guy an Audiofire12!

Thanks for doing all this, man!

[TAFKAT]

Originally Posted by zephonic

Any chance you can get your hands on an MR816, Vin? It would be interesting to see how it compares to the MBox3Pro.

I presume you used Cubendo for the plug-in counts and ms latency test?

Hey Z,

I'm on it , the MR's are prohibitively expensive here in OZ so not very popular, thats the only reason I hadn't tested one up until now. I finally managed to secure an MR816 CSX which I have tested up and will post the results shortly, as well as results for the AVID HD Native card under ASIO and the SSL Nucleus USB2 audio component.

All the tests are on Cubase 6 , which is pretty good at reporting the actual latency , at least for those interfaces that the ASIO driver is reporting the correct value to the app.

Re the Echo Audiofire , I may be able to get a hold of a test unit.

[TAFKAT]

O.K, here are the final results for the current series of testing.



Some notes on the latest interfaces tested.

RME HDSPe AIO - 3.024 MC On/Off - After the initial testing found that the 3.24 driver was not performing as well as the earlier 3.085 , it was suggested by M.C of RME that the new Optimize Multiclient Mixing option could be the variable in play. Sure enough when the option is disabled the performance was identical to the earlier driver. I have listed the RME HDSPe AIO with both the option On and Off.

The MOTU 424 PCI / 24 I/O delivered exceptional performance right down to 032 samples , and despite the overall numbers being below that of the RME reference card, if you take a closer look the I / O and RTL are measurably lower as the MOTU uses very minimal padding on the buffers. By taking that into account using the RTL % rating of 1.17 , it drew level with the reference RME HDSPe.

The MOTU's have always drawn mixed reports on Windows and I know first hand how flaky the cards can be having 3 different revs of 424 card across the bench in the last few weeks, and its a crap shoot whether the respective revs will work on current systems , but the drivers themselves have always been good with the PCI/PCIe cards, ( as long as you don't use external clocking- LOL )

The ESI U46XL was the USB audio interface that I had hinted at a while back that surprised me and the results speak for themselves. Yes the reported latencies are nominal and it has the oddity of having the highest buffer setting of 256 ( actually one step higher to 288 ) , so I had to do some juggling with calculating % ratings , but seeing that this whole exercise is about LLP , I gave ESI some rope with the interface not having a 512 setting. Great performing little interface , especially for it being USB2.

Staying with ESI, the Maya 44e is a mixed bag, respectable performance although beaten by the U46 , but the card has an issue where the driver has to be reset in Cubendo every time you open a project with any buffer setting below 512 ??? !!

PITA - Still waiting to hear back from ESI who have initially said they can not reproduce it.

The Steinberg MR816 delivered decent overall performance, latency settings available all the way down to 032 samples, mind you that's with the help of some extra padding on the output buffers, which resulted in higher RTL values and subsequent lower comparative RTL %. The 032 setting is more Window dressing IMO.

SSL Nucleus - the results don't tell the whole story as the unit is so inconsistent across multiple systems and even respective USB ports. The results are pretty decent but it took some major arm wrestling to settle the system to achieve those results. The driver panel still looks like the vanilla OEM breadboard and the odd non standard buffer sizes are a sore point if you want to try and use it with Protools 9 - which it is advertised as being compatible with - as PT9 simply will not accept the odd buffers.

The Mackie Onyx actually performed O.K for a Mixer/FW combo, DAWbench DSP were quite good, but it collapsed under the VI testing - the OEM controller definately feels different to some of the Dice variations, so not sure what Mackie are using there.

The Midas Venice was an interesting one, basically a mid size live desk with a Dice FW hacked into the back - I/O - RTL were identical and performance was close enough to identical to the Focusrite Sapphire , so its obviously using the same ( poor performing ) controller.

AVID Mbox 3 is identical to the M-Audio FTU range with slightly better I/O and RTL , not too hard to see that AVID are using the same under bonnet components across the AVID Mbox 3 Pro / Mbox 3 / Profire / FTU ranges.

AVID HD Native card under ASIO , well to say it was an interesting experience is an understatement. Input buffers are double the usual, Output are reporting Nominal , overall performance isn't too bad right down to 032 samples, but there were some major hoops to activate the ASIO driver in Cubase . i.e : needing to toggle off the ASIO driver back to the Generic and then back again to simply wake the driver on every session load .

Its obvious to me that the card is specifically geared to Protools , the ASIO driver really being only for convenience for some compatibility with other DAW hosts , so its not really a player in the current comparative.

Thats it until the next round..

Peace

[jcschild]

well done

[zephonic]

Wow, that MBox3Pro kicks hiney!

[Ivorydom]

Hey Vin,

This is an astonishing contribution, thank you so much. Some quite surprising results there.

Do you think that you will add the UFX soon? You wrote that the performance is a little behind FF800 in all cases, can you elaborate on that? I wonder why it would be behind FF800 especially on FireWire.

[TAFKAT]

Originally Posted by Ivorydom

Hey Vin,

This is an astonishing contribution, thank you so much. Some quite surprising results there.

Do you think that you will add the UFX soon? You wrote that the performance is a little behind FF800 in all cases, can you elaborate on that? I wonder why it would be behind FF800 especially on FireWire.

Hey ,

Thanks Mate, definitely a few cats amongst the pigeons there, and I'm sure a few more as this progresses.

What I am sharing publicly is my own internal data base as I qualify every interface that comes across my bench so that I know exactly where the bodies are buried for the respective clients systems. With some of the huge variables I was navigating with some of the Dice based FW interfaces and the lack of clarity from the manufacturers and so called product reviews, I thought it was time to get some focus on the LLP aspect and to maintain the focus so that end users know exactly what to expect when weighing up options for their particular working environments.

I am sure some of the manufacturers are less than impressed this is being aired, but you know what , they can find my contact details easy enough if they want to discuss it... :-)

Re the UFX testing, I will have access to a unit in the next few weeks so I'll run thru the suite across both the FW and USB2 aspects and post the results.

The performance of the UFX under USB2 I expect to be identical to the Babyface as they share the same under bonnet genealogy , however the FW aspect of the UFX has nothing in common to the earlier FW400/800 units as they had a dedicated FW controller ( TI ) that RME custom implemented , whereas on the UFX , all protocol arbitration is handled via the FPGA. I am not expecting it to be as tight , but am very interested in how the numbers pan out. The UFX IMO is specifically focused as a USB2 interface and it performs admirably under that protocol , I won't be overly surprised if the FW performance is actually below that of USB2.

[TAFKAT]

By popular demand , the LLP results in graphical form




* Quick Note - The AVID Mbox Pro 3 results and rating have been amended as I discovered I had a slight miscalculation in the RTL % results which has been amended from 1.04 to 1.05 , not huge but enough to bump the rating a notch. The tables have been amended as well *

[jeronimo]

Originally Posted by TAFKAT

By popular demand , the LLP results in graphical form




* Quick Note - The AVID Mbox Pro 3 results and rating have been amended as I discovered I had a slight miscalculation in the RTL % results which has been amended from 1.04 to 1.05 , not huge but enough to bump the rating a notch. The tables have been amended as well *

What is this MC on/off on the HDSP?


Posted via the Gearslutz iPhone app

[TAFKAT]

Originally Posted by jeronimo

What is this MC on/off on the HDSP?

Multi Client Mixing Option - I noted that on the charts a few post back.. :-)

I'll add an extra note in the comments

[zephonic]

Originally Posted by TAFKAT

* Quick Note - The AVID Mbox Pro 3 results and rating have been amended as I discovered I had a slight miscalculation in the RTL % results which has been amended from 1.04 to 1.05 , not huge but enough to bump the rating a notch. The tables have been amended as well *

Could you elaborate on the MBoxPro3's general performance a little more? From the graphs it looks like a step up from the other FW devices on test, but could you briefly describe how it fared subjectively compared to -say- the MR816?

And just to make sure: RXC is what? And CV and NCV are Kontakt with Convolution Reverb and without?

Thanks.

PS: Roland touts their Octa-Capture (and Quad-Capture) as having extreme low-latency drivers (VS streaming is what they call it). Hope you can get your hands on one of those soon.

[Jomox]

Would still love to see the EMU 1616m in the mix (1820m would do though don't have as good pre amps, circuitry or drivers)

Just interested to see if it's worth upgrading and how well my 1820m roughly performs compared to the competition (1616m performance would be plenty)

What do you need to get hold of a 1616m to do the test?

[remo]

Hey great thread. I'm happy to drop over my RME Fireface UC if you want to test it, I'm also in Melbourne. Shoot me an email guy at nixonaudio dot com.

[troggg]

Tafkat, can you be more specific about what exactly you consider low latency performance in the real world?

I sort of know, everyone sort of knows, but hearing exactly what you consider it to be would make your published results even more valuable.

I'd be interested to see some of you guys use this app with your host:
Game Prelauncher - Ultimate Optimizer for Games - official site

and see how it affects latency.
SHould make a significant difference.

[TAFKAT]

Originally Posted by zephonic

Could you elaborate on the MBoxPro3's general performance a little more? From the graphs it looks like a step up from the other FW devices on test, but could you briefly describe how it fared subjectively compared to -say- the MR816?

Hey Z,

The performance of the Mbox3 Pro was very solid , I was pleasantly surprised that not only was the I/O - RTL lower than the Profire 610 which shares the same controller , but also that the performance was better even with the lower I/O and RTL.

The AVID driver development team have really connected the dots with these new interfaces, they scaled extremely well right to the nth degree , were very consistent. The only other FW units in that league are the RME's.

As a head to head comparative against the MR816 purely on performance, the drivers felt way better on the AVID when pushed, and with substantially lower I/O and RTL

And just to make sure: RXC is what? And CV and NCV are Kontakt with Convolution Reverb and without?

RXC - ReaFX - ReaXComp - listed at the botton of the table BTW.. :-)

CV and NVC , yep CV has Convoverb on the Orchestral elements, NCV has No Convoverb on those respective parts.

Not a lot of difference as you can see except at ultra low latency, where the CV and NCV sessions become interesting is where my current system cannot run the CV version at 032 samples , the faster Quads and Hexacores deliver measurable improvement which will hold us in good stead as the newer tech rolls out later this year , so we can do some head to head.

PS: Roland touts their Octa-Capture (and Quad-Capture) as having extreme low-latency drivers (VS streaming is what they call it). Hope you can get your hands on one of those soon.

Hmmm, I still have a few good contacts at Roland from a past life, I'll ask the question.. :-)

@ Jomox,

Don't really have any avenue to approach the local reps unfortunately , so unless an end user sends me one to place under the microscope, I can't see me testing the EMU's unfortunately.

Originally Posted by troggg

Tafkat, can you be more specific about what exactly you consider low latency performance in the real world?

I sort of know, everyone sort of knows, but hearing exactly what you consider it to be would make your published results even more valuable.

Real World is very different for each individual and its horse for courses in many repsects. A lot of my clients demand solid working environments at 064 samples not only for realtime playing of Virtual Instruments , but also for monitoring with FX and using Guitar Amp Sims for example. IMO on the current multicore systems ( especially since Nehalem ) 064/128 is realistically where you should be able to set and forget for anything except final mixing stages , and this is where the efficiency of the audio interface drivers can have a huge influence on what you will actually achieve performance/scaling wise.

256 is right on the edge for many in regards to real time playing of VI's and as you can see from the charts, even at that level which I would consider moderate latency , some interfaces fall in an absolute heap when using sample based VI's , which is more then a little concerning if you happen to be using one of those interfaces and lose 40-50% of your potential overhead due to a ill performing driver/controller.

FWIW: I'll share some Real World feedback I received from one of my Nuendo Post clients this week who had a renown 3rd party come in to do some final mixing on a Film - 300 odd tracks of audio , edits, fades, automation, etc, tons of top end native plugs , Waves, URS, etc and the engineer was just staggered that this system was running rock solid with the ASIO meter starting to teeter into the higher 80% ASIO loadings . He was even more staggered when the client who owned the system checked the buffer setting on the RME HDSPe AES card , and it was still on his set and forget value of 064.

I have known this system to roll down to 032 samples to feed really tight vanity effects to some particularly finicky/demanding talent, and for the system to be accidentally left there for days on end without issue

The system if anyone is wondering - X58 / Xeon 3550 - 3.06 GHZ / 12GB DDR3 , so basically identical to my dev system bar some extra clock.

@Remo,

Thanks for the offer, I'll drop you an email and take you up on that when I get some clear air.. :-)

[clonewar]

This is really great info, thanks for taking the time to test and share the results Vin!

One question.. The RME HDSPe AIO is your baseline, do you know how the RME PCI cards, like the 9632, stack up against it in these tests?

[TAFKAT]

Originally Posted by clonewar

This is really great info, thanks for taking the time to test and share the results Vin!

One question.. The RME HDSPe AIO is your baseline, do you know how the RME PCI cards, like the 9632, stack up against it in these tests?

Hey ,

The performance of the HDSP 9632 ( my old reference card ) and HDSPe AIO were identical in the tests I conducted on the X58 dev system when I changed over.

@ Soulstudio,

Very interesting, definitely worth a look see, thanks for the heads up.. :-)