Audition 2 and ASIO - problems

Hmmm - not short of 'grunt' there then.  Did you try the Asio4All driver?  I forget.

Not short of grunt there, then.  

Are the SATA ports in the motherboard chipset, or are they additional?  There are still motherboards around where the SATA is added on via the PCI bus, and this can cause PCI latency problems with many sound cards.  I wouldn't have thought this would be a problem with an 64-bit mobo, though, but it's worth a check.  Sometimes it's worth adjusting latencies anyway, using the (free) tool from note 26 here.

Paul

Not short of grunt there, then.  

Are the SATA ports in the motherboard chipset, or are they additional?  There are still motherboards around where the SATA is added on via the PCI bus, and this can cause PCI latency problems with many sound cards.  I wouldn't have thought this would be a problem with an 64-bit mobo, though, but it's worth a check.  Sometimes it's worth adjusting latencies anyway, using the (free) tool from note 26 here.

Paul

The SATA ports are integrated into the motherboard's Nvidia chipset.  Ill give ASIO4ALL a try, although i am skeptical about its benefits.  I'll also try the PCI Latency tool, it seems to have been useful for me in the past.

ASIO4ALL caused audition to crash immediately, and changing the PCI Latency is ineffective.

More hmmmm.... the program is entirely happy with asio4all here, so why not there??

Just in case, have a rummage around Adobe's tech notes - it's surprising what's in there.

Well, I fired up Sonar 4 to see if it had any ASIO problems like Audition 2.0.  I piled on tons of realtime effects on a dozen 96khz tracks, running my card's ASIO driver @ 256 sample latency, and there is absolutely no problem at all, the sound is perfect.  

I am thinking that the problem is now an Audition problem.  

Edit:  I did read the tech notes and system optimizations for Audition, as well.  They didnt help whatsoever with the problem.

You do have lots of effects in that session.  Echo Chamber uses lots of cpu power.  When you add it, I think it actually warns you that it's not suitable for real time (especially if you have more than 1 or if you have other effects in other tracks).

Audition 1.5 had background processing.  So at all time, it was mixing down tracks down to a single stereo track.  When you pressed play, you were playing back a single (background mixed) stereo track.
Audition 2.0 doesn't have background mixing (because we added ASIO, low latency, etc). So when you press play, you are actually playing back all the individule wave files simultaniously. So now you are limited by hardware.  Also, in audition 1.5, we were not as concerned with low latency and our effects weren't optimized for that.  In 2.0, when we optimized our effect for MT low latency, this reduced the number of effects you can add to the session (mainly for heavy cpu effects).  this is because we now have to process smaller chuncks of data to produce low latency.  reading lots of small chunk of data (but alot more of them) reduces the number of real time effects you can have (limited by your Hardware).

If you have large session with several effects or medium sessions with lots of real time effects on them, in 2.0 you might reach your hardware limitations sooner than you did in 1.5.  I recommend that you freeze / Lock tracks with effects.  

Here is another solutions you are not going to like (since you have AMD).  We did optimize audition 2.0 to work with SSE processors (which you have).  But we also added some optimizations for Intel chips.  you may see a better improvement if you use Intel P4 processors.

I remember that when we added SSE support NR effect was twice faster.  When we added intel optimization, we again improved the NR processing by something like 50%.  This was only effect I remember checking out.  This improvment wasn't as obvious on other effects. And that improvement was in Edit View only.  I didn't check it in MT view.  So P4 may give you a bit more improvement.  

Another possible solution:  I recall that we had a 1.5 session that played great in 1.5 and really bad in 2.0.  It had 30 tracks and bunch of effects on about 20 tracks.  Even freezing the tracks didn't totally solve the problem (though it improved it greatly).  It took us a while, but we finally found a solution.  Defrag.  Simply defragging the drive on that machine solved the problem.  I'm not an expert in HD read/write/access time, but I'll try to explain what I undertand in this case. Audition 1.5 uses large chunk data to read / write to / from HD.  Since 2.0 is low latency and it reads (access) data from HD more often, it make a big difference if your HD is defragged or not.  And remember now that audition isn't doing background mix, it's reading lots more info (tracks) from HD than it did before. So it is very important to defrag your HD often.  Windows defrag solved the problem for our session stuttering.  But I heard there are better defrag tools that do better job.  Maybe someone on this forum can recommend some they have used.

I hope this info helps explain how audition 2.0 works differently and one of these tips helps solve or improve playback of your session.

Thank you for the reply.  It sheds much light on the situation.

Here is another solutions you are not going to like (since you have AMD). We did optimize audition 2.0 to work with SSE processors (which you have). But we also added some optimizations for Intel chips. you may see a better improvement if you use Intel P4 processors.

I definitely am concerned about that.  AMD processors are currently faster than Intel's offerings in virtually every outlet of computer processing, consume less power and have a better price/performance ratio.  It must be a whole lot easier to optimize for Intels, otherwise I have no idea why else Adobe would not optimize for AMDs as well.  

Defrag.

I didn't see a need to, because I had completely reformatted the drives and installed Windows XP a few days ago, and the defrag program said I did not need to defrag.  I did defrag the drives anyway, but it did not make any difference in Audition's performance.

If you have large session with several effects or medium sessions with lots of real time effects on them, in 2.0 you might reach your hardware limitations sooner than you did in 1.5. I recommend that you freeze / Lock tracks with effects.

I was afraid that might be the only choice I could make, but as I said earlier:

It just doesn't seem right to me that AA2.0 can't even run 3x Dynamics Processing and 2x Graphics EQ realtime @96khz w/ a 512 sample buffer size without popping or glitches on my machine.

And I am very skeptical that it is a hardware limitation, proof being that I can run a very large number of realtime effects in Sonar 4 without a single hitch, and

...when playing back the session in [Adobe Audition] 2.0, although the audio crackles and pops horribly at low ASIO latencies, cpu usage doesn't go above 64%

Therefore, I must conclude that it is not a hardware limitation causing the problem.  It seems to point to Audition being grossly unoptimized for AMD machines, and/or having an unoptimized implementation of ASIO.    


This is very upsetting, to say the least.  I can only hope that Adobe investigates and releases a patch, but I haven't seen others complain about this problem (yet), so I guess i am SOL for the time being.

Interesting stuff...

Given the number of freeware VSTs out there it should be possible to compare one DAW with another more accurately using VST effects only.  The "Ambience" reverb springs to mind.  I think it is fairly demanding.  I'll try to put together a test which can readily be replicated elsewhere to compare AA2.0 installations one with another and vs other DAWs.

that seems like a good idea, so I did a quick test.

Testing @96khz, 128 Sample Latency

I took four drum tracks and loaded them into Audition, and placed Glaceverb (a free reverb vst) over each track (using the "Classic - large Theater" preset on each).  No pops or crackles.  I add one more Glaceverb to any track, pops and crackles.

I took the exact same tracks and loaded them into Sonar 4, placed Glaceverb over each track.  No pops or crackles.  I then was able to add 3 more to the session before it started to have problems.

The Summary:

4 Glaceverbs in Audition glitch-free.
7 Glaceverbs in Sonar 4 glitch-free.

Sonar is 43% more efficient in this test.

Edit: Inserted Link

Hmm, trouble with better-than-CD quality audio for testing is that it's not so easy to replicate generally - I don't have any for a start!  With CD quality you can quickly rip some tracks from a CD, insert each song on a MT track, and test.

But using the freeware VST "Ambience" reverb, "Amoebe Hall" preset, AA1.5 will play 5 tracks glitch free with mixahead disabled - AA2.0 manages 3 tracks with 512 sample latency.  (That is using CD audio).  It won't play 5 tracks with any driver nor any latency, ASIO or not.

When it comes to mixdown, AA2.0 bounces down two minutes-worth of five Ambience-reverbed tracks in 1'57" - or freezes the first track in 52"

AA1.5 bounces them in 1'46" and freezes track one in 44".

Nariman's point is partly that to provide the functionality users were asking for, some of the efficiency has been lost.  I think also one has to bear in mind that this is a substantial - even total? - re-write of the code, whereas up till AA1.5 it had been honed over many years from a much more lightweight start - comparison with the efficiency of other apps that have been tuning their existing code base for some time is likely to produce unfavourable comparisons.

This version is going to work best with the use of buses for effects carefully optimised.  Here I can record and play back plenty of "plain" tracks at once - it doesn't seem to be the disk i/o which is the problem.

In the next version, it would be good to see not only a general tweaking of performance, but also some improved techniques used if possible - for example, if you bounce to a new track and solo that track in order to get around performance issues, the engine still processes the unplayed tracks, so the single solo track still glitches if the effects on the non-played tracks are intensive.  That seems worth revisiting.  Also, I'm pretty sure that Tracktion only processes those parts of tracks which have clips on them.  Audition has always processed the whole track, applying effects to silent portions, and that seems to be the case with this version too.  A few tweaks like that would add to any general increase in efficiency which it might be possible to provide down the track.

Are you using your card's ASIO driver?

If you are using ASIO4ALL instead, AFAIK you are essentially wrapping the WDM drivers in an ASIO package, not really using your hardware's ASIO features.  IIRC, you said that it has worked better for you, but it just doesn't seem like it should...

and even based solely on your findings using AA2.0, don't you find

Optimized performance

Experience dramatic performance upgrades — up to three times faster for some features — and optimized use of hard-drive space.

misleading when referring to realtime effects processing?

Somewhere the features have been listed which fulfil that speed claim.  It's not an untrue claim, but one which might readily be misunderstood.  

The ASIO driver in use seemed to make little odds.  As mentioned elsewhere, the Asio4All driver produces markedly better metering here, so I'd think it's more than just a wrapper.  Dunno, too technical for me.  I tried the non-asio driver and it was dire!

I loaded up my worst session ever made in AA1.5 and have an illustration of what Nariman was explaining.  It very nearly chokes my computer in AA1.5, but given enough starts and stops the background mixing does its job and you can work with it eventually, albeit with some delays and stutters.  In AA2.0, I can't -- all choking, all the time.  This creates quite a quandry, because had I created this session in AA2.0, the mix would have been finessed much more quickly because of the new features, particularly automation.  Greatly preferable!  But the scope of this session, even with locked tracks, is beyond the HDD's and CPU's ability to compute in real time.

So I would have to say there are three options:  bouncing down track groups and eliminating those tracks from the mix load; asking my employer for a computer upgrade, an approach that's not very practical for home users; or use AA1.5 for really huge sessions but lose all the benefits of the upgrade.  Wow.

Somewhere the features have been listed which fulfil that speed claim. It's not an untrue claim, but one which might readily be misunderstood.

It is not an untrue claim, but it is misleading due to its ambiguosity in refraining from describing exactly what parts are "up to three times faster", or providing a direct link.  Clicking the more features link at the bottom, though, does reveal that the optimizations (only?) benefit features in the edit view.  It is also misleading in a sense because we have seen that many features are actually much slower (pertaining to realtime effects) than 1.5.  Sonar 4 shows that ASIO implementation does not need to be.

I loaded up my worst session ever made in AA1.5 and have an illustration of what Nariman was explaining.  It very nearly chokes my computer in AA1.5, but given enough starts and stops the background mixing does its job and you can work with it eventually, albeit with some delays and stutters.  In AA2.0, I can't -- all choking, all the time.  This creates quite a quandry, because had I created this session in AA2.0, the mix would have been finessed much more quickly because of the new features, particularly automation.  Greatly preferable!  But the scope of this session, even with locked tracks, is beyond the HDD's and CPU's ability to compute in real time.

So I would have to say there are three options:  bouncing down track groups and eliminating those tracks from the mix load; asking my employer for a computer upgrade, an approach that's not very practical for home users; or use AA1.5 for really huge sessions but lose all the benefits of the upgrade.  Wow.

If you read my previous posts, you know that even though I have a computer that pretty much "top of the line", audition 2 is still very slow when using realtime effects.  

I had already came to that conclusion (many posts back) that I will need to heavily change my work habits in 2.0 (meaning freezing practically all my realtime effects), which is a large hindrance to start, but could be beneficial in cleaning up my workflow in future work.  

What I am questioning now is why Sonar 4 utilized ASIO in a much more efficient way (as I said before, I could put a very large amount of realtime effects in my mixes without a single hiccup), whereas AA 2.0 chokes when applying only a few realtime effects.

With VST effects, Sonar is evidently more optimized (my one test showing 43% more optimized, although clearly not representative).  

And with me throwing on many realtime dynamics compressors, reverbs, and gEQs, Sonar's internal effects seem to be much better optimized than AA 2.0's.