More Latency...SMAART measurements

[hclague]

Hi All,

Thought I'd start a new thread with these actual latency measurements of SAC, RME Raydat, and the Behringer DCX. Moderators feel free to combine if you like.

The Measurement setup:

SMAART Live 5 running on a toshiba laptop with M-audio mobile Pre. SMAART sends pink noise to an analogue mixer channel. Main out of analogue mixer feeds into SAC, output of SAC feeds the test signal side of SMAART. Aux 1 of analogue mixer feeds reference channel in SMAART.

Pic 1. Is A Control measurement shot the test signal is not running through SAC yet is feeding directly out of the analogue main out. Therefore there should be virtually no latency.( .02ms)
679

Pic 2. is the test signal run through ADA8000 into SAC via the RayDat card set at 2x64 and back out the ADA8000. 1 Channel loaded with EQ and Comp applied. (7.48ms)
680

Pic 3 is the test signal run through a Focusrite Saphire Pro40 into SAC via the RAyDat card set at 2x64 and out the Raydat card via AES/EBU into the Behringer DCX2496 system controller and then out the mid send of the DCX with no delay applied in the DCX. 21 Channels loaded with EQ and Comp applied. (8.44ms)
681

Pic 4 is the test signal run through a Focusrite Saphire Pro40 into SAC via the RAyDat card set at 2x32 and out the Raydat card via AES/EBU into the Behringer DCX2496 system controller and then out the mid send of the DCX with no delay applied in the DCX. 21 Channels loaded with EQ and Comp applied. (5.71ms)
682

Pic 5 is the test signal run through a Focusrite Saphire Pro40 into SAC via the RAyDat card set at 2x64 and out the Raydat card back into the Saphire Pro40 out the Saphire Pro40's analogue outs. 21 Channels loaded with EQ and Comp applied.(6.81ms)
684

Some conclusions:

1. Different Preamp/ADA converters have slightly different latency. No big revelations here.

2. Adding CPU load ( more channels EQ's Comps ) does not seem to change the throughput latency much.

3. Adding Digital system controllers will add between 1 to 3ms additonal latency even with no delay applied in the unit. Again, no big revelations

4. Going from 2x 64 to 2x32 on the RME Cards makes (what I consider) a significant difference.

[gdougherty]

Nice, so scenario 2 is what I'd measure in to out using SAC as mixer and speaker processor at 2x64?

Don't forget the CPU penalty of jumping to 32 sample buffers. The latency is nicer, but buying 2ms of latency with a 20% bump in CPU load on the 3GHz Athlon X2 we use at my church doesn't seem very practical to me when we're already running 30-50% loads depending on the channel use.

[AndyW69]

I was running 4x32 buffers and getting about 20 slipped buffers over a 6 hour rehearsal and show, I have switched to 2x32 (both on a RME pci madi card) and am seeing no slipped buffers, cpu load (8500) only a little higher, latency seems great (me musicians and performers have not noticed)

[DominicPerry]

Hal, are you able to run a test at 1x32 with RayDAT and ADA8000s and no DCX in the signal path? And could you do it at 88.2KHz as well as 44.1KHz? That will give something approximating the 'best' SAC can do, bearing in mind that very fast convertors like RME-ADI8-QS are probably impractically expensive for just about everyone.

Dominic

[ssrsound]

ADA8000 can only do 48 maximum. So you'd have to change your preamps to do that test.

[Brent Evans]

Interesting. I didn't imagine there would be as much as 1ms difference between two different RME cards with identical settings.

I wonder how much the chipset/processor combination actually comes into play? What were the specs on the test machine?

I'd also be interested in seeing some tests run on MOTU hardware, I'm considering that setup for a portable system.

Good info, great follow up, thanks!

[DominicPerry]

Interesting. I didn't imagine there would be as much as 1ms difference between two different RME cards with identical settings.

I wonder how much the chipset/processor combination actually comes into play? What were the specs on the test machine?

I'd also be interested in seeing some tests run on MOTU hardware, I'm considering that setup for a portable system.

Good info, great follow up, thanks!

Watch out with getting the right MOTU products. Plenty of MOTU-Misery around the forum.:eek:

Dominic

[DominicPerry]

ADA8000 can only do 48 maximum. So you'd have to change your preamps to do that test.

Good point. Maybe with the Saffire Pro 40 then?

Dominic

[Brent Evans]

Watch out with getting the right MOTU products. Plenty of MOTU-Misery around the forum.:eek:

Dominic

Oh I know. The only one I'd consider would be the 324/424-2408 combo.

[hclague]

Nice, so scenario 2 is what I'd measure in to out using SAC as mixer and speaker processor at 2x64?

Don't forget the CPU penalty of jumping to 32 sample buffers. The latency is nicer, but buying 2ms of latency with a 20% bump in CPU load on the 3GHz Athlon X2 we use at my church doesn't seem very practical to me when we're already running 30-50% loads depending on the channel use.

Hi George

I guess I wouldn't guarrantee that adding system control filters ( Rubber filter etc..) wouldn't add more latency since I didn't have those in the system, but other plugins ( that work in SAC ) don't seem to add much if any.

Yes. CPU load does jump up with 2x32. Mine went from 45-50% to 65-70%, but I would like to stay as close to 5ms as possible.

Hal

[hclague]

Hal, are you able to run a test at 1x32 with RayDAT and ADA8000s and no DCX in the signal path? And could you do it at 88.2KHz as well as 44.1KHz? That will give something approximating the 'best' SAC can do, bearing in mind that very fast convertors like RME-ADI8-QS are probably impractically expensive for just about everyone.

Dominic

Hi Dominic

I will try to take a look at this. i imagine the slipped buffers will not make a difference in the latency measurement?

Hal

[hclague]

I wonder how much the chipset/processor combination actually comes into play? What were the specs on the test machine?


Hi Brent

The test machine was:

Asus P5Q-SE2 Motherboard
4 gig of DDR2 1066 memory
E8500 Processor

Hal

[DominicPerry]

Hi Dominic

I will try to take a look at this. i imagine the slipped buffers will not make a difference in the latency measurement?

Hal

I wouldn't have thought so.

Thanks Hal.

Dominic

[Brent Evans]

Hi Brent

The test machine was:

Asus P5Q-SE2 Motherboard
4 gig of DDR2 1066 memory
E8500 Processor

Hal

Interesting. Your machine is faster and ran less processing than mine (I just patched in to my normal template, lots of processing going on). So the only difference between your test and mine (other than the complexity of the testing equipment) was the sound card. The biggest difference between my DIGI and your HDSP card (at the same buffer size) is TotalMix. Did you have Totalmix enabled? If so, is it possible to run the test without Totalmix and see if it shaves that ms?

[DominicPerry]

You can't turn off TotalMix with the HDSP cards.
It may be that the HDSP cards are simply faster than the older generation. Or are you saying it's slower?

Dominic

[Brent Evans]

You can't turn off TotalMix with the HDSP cards.
It may be that the HDSP cards are simply faster than the older generation. Or are you saying it's slower?

Dominic

His HDSP 2x64 was 7.5 ms, my Digi was 6.5 ms. It's not that big of a difference, but it is a difference.

[DominicPerry]

His HDSP 2x64 was 7.5 ms, my Digi was 6.5 ms. It's not that big of a difference, but it is a difference.

Thanks Brent. I was simply too lazy to go and find your results. Sorry.

You're probably right that TotalMix accounts for the difference. But that's just a guess.

Dominic

[Brent Evans]

Thanks Brent. I was simply too lazy to go and find your results. Sorry.

You're probably right that TotalMix accounts for the difference. But that's just a guess.

Dominic

There's another test to run too.. patch a channel directly through TotalMix and compare that to a signal passed through SAC to find out exactly how much latency SAC imparts.

[hclague]

You can't turn off TotalMix with the HDSP cards.
It may be that the HDSP cards are simply faster than the older generation. Or are you saying it's slower?

Dominic

I believe you can turn off total mix. I'll try it tonight.

Hal

[hclague]

So the only difference between your test and mine (other than the complexity of the testing equipment) was the sound card. The biggest difference between my DIGI and your HDSP card (at the same buffer size) is TotalMix. Did you have Totalmix enabled? If so, is it possible to run the test without Totalmix and see if it shaves that ms?

"(other than the complexity of the testing equipment)". This could also account for the difference ( not apples to apples ) but I'll see if I can turn off total mix and re-measure. Otherwise i will try to determine the latency of Total mix. It would be discouraging if Total mix is adding to latency and cannot be turned off if one is not using it.:(

Hal

[905shmick]

Make sure performance settings in windows has been changed from 'Applications' to 'Background tasks'.

[Brent Evans]

"(other than the complexity of the testing equipment)". This could also account for the difference ( not apples to apples )

I'd accept a small margin of error on that point, and my wav files are published for analysis, but that's almost a 15% difference, which is statistically significant.

The results will be interesting, to say the least.

[hclague]

Update.

I turned off the HDSP mixer last night, and re-measured. There was no change in latency. This makes sense to me since I don't believe I was running anything through the HDSP mixer. It looks like all the latency is derived by the Buffer settings and ADA8K ( or Saphire Pro and/or DCX ).

I tried to switch to 88.2k sample rate but SAC would not open and gave me an error pop up. Probably because the interfaces can't do that sample rate?

Hal

[gdougherty]

I've found on my rackmount rig with HDSPe RayDAT that the system will not pass audio until I've started TotalMix.

[Jeff Scott]

The almost 8ms of latency from Pic 2 (Raydat thru ADA8000) may be a deal breaker when combined with Digital drums (seeing that they are adding several ms of their own). As I stated in another thread...my drummer can still hear lthe latency at 2x64 though he admitted it's much better.

I will be playing at a local club this Saturday...with Digital drums and with SAC. I'm going to split his Drum feed so that he is not having to deal with the 8ms+ that SAC is adding. My main concern is the delay coming out of FOH: will it be noticable? I'll let you know.:(:confused::eek:

[Bob L]

I am not so sure the numbers and tests are perfectly accurate... but...

Trust that any latency coming out floor wedges or FOH is just not an issue... any speaker on the floor or in the air is already generally more than 10 ft away and already adding 10ms or more of latency and no one has ever complained... you can see we are doing major shows and events on a daily basis with no conversation concerning latency... and to be honest, no issues with any in-ear performers... drummers and vocalists included.

So... I suggest to get out of your over-thinking and testing mode and simply patch it in and use it and listen and enjoy... the ride is quite thrilling. :)

Bob L

[gdougherty]

The almost 8ms of latency from Pic 2 (Raydat thru ADA8000) may be a deal breaker when combined with Digital drums (seeing that they are adding several ms of their own). As I stated in another thread...my drummer can still hear lthe latency at 2x64 though he admitted it's much better.

I will be playing at a local club this Saturday...with Digital drums and with SAC. I'm going to split his Drum feed so that he is not having to deal with the 8ms+ that SAC is adding. My main concern is the delay coming out of FOH: will it be noticable? I'll let you know.:(:confused::eek:

We use ears and a Roland TD-6 on a frequent basis with no complaints from guest drummers.

[Wink0r]

As far as front of house goes there are some engineers that delay the FOH (induced latency) to move the speakers back to the plane of the backline. Latency in the FOH only has the effect of moving the stacks back by roughly one foot per millisecond.

[KUI]

Try rerunning the test with your generator source hard wired to the reference input, that will take the analog desk out of the equation.

kui

[Ogmeister]

As far as front of house goes there are some engineers that delay the FOH (induced latency) to move the speakers back to the plane of the backline. Latency in the FOH only has the effect of moving the stacks back by roughly one foot per millisecond.

I do this all the time. This helps drummers a lot especially on smaller stages with keeping the bass guitar amp and drums in time alignment with the house stacks. When I get this right the drummer will not need to much low end in the
monitor as the house subs will give him a nice thump. In some cases if I have a low end problem on the stage and moving the house subs is not an option I can put on delay to help with the hump on stage. I use delay and phase frequently on monitor wedges to correct the hot spots of feedback.

Since I have been using SAC I just love the control I have on the monitor mixes. I can get loud and clear monitors with great sound. Something that takes a lot of outboard tricks to get working on my analogue monitor boards. I don't miss those outboard racks one bit.




OGO:)

[Jeff Scott]

I do this all the time. This helps drummers a lot especially on smaller stages with keeping the bass guitar amp and drums in time alignment with the house stacks. When I get this right the drummer will not need to much low end in the

. I use delay and phase frequently on monitor wedges to correct the hot spots of feedback.

Since I have been using SAC I just love the control I have on the monitor mixes. I can get loud and clear monitors with great sound. Something that takes a lot of outboard tricks to get working on my analogue monitor boards. I don't miss those outboard racks one bit.

OGO:)

Og: Can you elabortate a bit more on the FOH and the monitor tecnique? For delaying FOH for example: if your front FOH stacks are 10ft in front of your backline / Drummer..do you delay the FOH the equivalent of that 10Ft?

Please explaine using Phase and delay to help with Monitor feedback....

[Ogmeister]

Og: Can you elabortate a bit more on the FOH and the monitor tecnique? For delaying FOH for example: if your front FOH stacks are 10ft in front of your backline / Drummer..do you delay the FOH the equivalent of that 10Ft?

Please explaine using Phase and delay to help with Monitor feedback....

Yes this delay would equal the distance from the back-line to the front stacks.
As for phase and delay on monitors. Sometimes if I have a null or a hot spot at some point on the stage relating to another wedge or house system I might use some delay or phase swapping to correct. Especially if I don't have the option of moving physically the trouble maker. By the way; speaker placement is one of the most important things to do first when fixing feedback problems.
What I train my engineers to do is this: if you have more than 7 bands on an eq trying to fix a feedback then its very likely eq is not going to get you the result. It also helps to know the axis response of the mics and place the monitor speakers to take advantage of that. Don't try to deny the physics of sound it will win every time you battle it.
Learning what you can fix with eq and can't will save you a lot of grief.
Another important thing to learn is what all the octaves are and how the band instruments and vocals are placed in them.
Train you ears to detect what frequency or note is causing the feedback.
Put away those analyzers and learn to do this by ear. Once you take the time to master this you will be the king of feed-back control and the band will like you a lot and you will gets lots of repeat business.

OGO:)

[Wink0r]

Knowing the frequency of notes and how octaves work is important. Many times a musician can tell you a note name for a feedback frequency that he is hearing. He probably won't be able to tell you a frequency but if you can quickly convert his note to possible frequencies you will be way ahead.

[Brent Evans]

I am not so sure the numbers and tests are perfectly accurate... but...

Trust that any latency coming out floor wedges or FOH is just not an issue... any speaker on the floor or in the air is already generally more than 10 ft away and already adding 10ms or more of latency and no one has ever complained... you can see we are doing major shows and events on a daily basis with no conversation concerning latency... and to be honest, no issues with any in-ear performers... drummers and vocalists included.

So... I suggest to get out of your over-thinking and testing mode and simply patch it in and use it and listen and enjoy... the ride is quite thrilling. :)

Bob L

Respectfully, this is valuable information to know. When I first showed SAC to a touring pro audio guy I work with from time to time he had two questions: How many layers does it have (thinking about it like a digital mixer) and how much latency is in the system. He saw the benefit of the interface pretty quickly, and I gave him a number based on the sound card settings (I think i said 5ms, which isn't out of line), and he was impressed and satisfied.

Some people live and die by specs, regardless of how right or wrong that is. It's not that 6.5 or 7.5 ms is going to make a huge difference, to them it's knowing that's what it is, and then they're comfortable.

Further, some of us have fun measuring stuff. It's neat to do and just as satisfying as getting a mix right, or knowing you have the best hardware out there.

[Bob L]

But... in many cases the results can be flawed by faulty test conditions that are not immediately obvious.

We have done our own tests with completely different results... more back down in the 4.5ms to 5.5ms range.

The frequency of the triggered pulse can make a difference... lots of things can make a difference... that's why I suggest being careful to come to conclusions based on a number and use your ears in a real test situation.

Can you do real shows... how does it sound... is it stable enough to trust... etc... and without a doubt... we are doing it every night in major showrooms and events.

Bob L

[Brent Evans]

This is true. No one is saying it doesn't work, or that the tests are flawless. In fact, the findings of disparate results which are fairly close, but with a likely statistically significant difference, indicates that there are too many variables to lock this down completely. It's simply good to know a range of limits and capabilities.

I am curious about something, however. How would the frequency of a test tone change the measurement in any way that is statistically significant?

[KUI]

I believe that Bob was refering to the sampling frequency ie, 44.1K, 48K, etc.

kui

[Bob L]

To test this correctly you pretty much need to put a pulse tone of some sort thru the system and then measure the time it takes to receive that pulse on the other end... the rise time of the frequency of that pulse can significantly affect the results of where to take the measurement on the other end.

You most likely want to use some sort of squarewave pulse of a certain width and frequency in an attempt to keep the rise time as close to vertical as possible... you will find that the frequency and width of the selected trigger tone and the system components and room characteristics can affect the received pulse tone's risetime and shape significantly... thereby adding ambiguity to the calculated results depending on where you or some automatic trigger detector decide the waveform edge has been recived.

And yes... sampling frequency is most important and any published latency results from any system would need to include the samplerate... and in many cases that may conveniently be missing... which means the results may be quite misleading when comparing one system to another.... for instance... a 64 sample buffer at 44.1k would calculate out to 1.45 ms... but the same buffer at 48k would be 1.33 ms and at 96k it would only be .66 ms... when figuring final latency of x number of buffers, the final result would be quite different... and without comparing systems at the same rate, the conclusion could be quite false.

Bob L

[Ogmeister]

And of course you have the slew rate of any OP-Amps or other analog circuits.
Every circuit changes what it process (hopefully in a useful and musical way) you can take an analog device like a mixer or gate (whatever)and you can see some sort of change being impressed to the passing signal.

Bob.. Before all these fancy tools these guys have access too now we used to use our ears to figure out what sounds good. I have to say that your mixing software continues to amaze me every time I am out with it.
I only wish for the day when all of my riders are asking for it.

Its is all about the result for me and I don't try to figure out how you did it but you did it great!!!! We patiently await your latest version as we know it will be amazing!

You are the mixing KING!!

[Grekim]

A simple and accurate (IMHO) way to measure latency is put a mic near or in some headphones. Then maybe a foot or so away strike something with a simple transient like a pair of drum sticks or two pieces of metal....obviously something without any extra vibrations. The mic will pick up the the initial strike and the latency coming through the headphones. So you have one recorded track with all the info you need.

[Brent Evans]

A simple and accurate (IMHO) way to measure latency is put a mic near or in some headphones. Then maybe a foot or so away strike something with a simple transient like a pair of drum sticks or two pieces of metal....obviously something without any extra vibrations. The mic will pick up the the initial strike and the latency coming through the headphones. So you have one recorded track with all the info you need.

That's essentially what I did in my setup, except I captured in stereo and electronically instead of acoustically (for precision). I think the concern is that the shape of the waveform might prevent accurate measurements. I may repeat with a generated square wave when I get some time.