Audio drop out with SP Caen set

[mdyde]

Hello Chris,

> A polyphony of 3000 reaches 6000 when using the surround!

I would imagine that the Hauptwerk setting would remain the same, but the effective polyphony is halved?

Correct - each pipe of a surround set uses two voices of polyphony, so you'll be able to play half as many pipes with a given amount of computer power.

Conversely, you'd need twice the CPU power in order to be able to play the same number of pipes simultaneously. I.e. take the traditional CPU requirements and double them.

I don't currently have a copy of the Caen sample set, so I'm afraid I can't comment very reliably, but for a very large and wet surround-sound organ the size of the Caen (55 ranks?, i.e. equivalent to 110 stereo ranks) you might find you really want an 8-core computer to play its tutti.

I wonder if it is possible some of the release samples are incorrectly sized, thus reducing the polyphony further (to the point new notes can't start). I say this because when hitting the wall of silence, if I stop playing, it takes a while for CPU useage to go down again.

Presumably the release tails are extremely long, which would definitely have a big affect on CPU requirements. Maybe Jiri can advise on that.

I also get this with my Caen with the surround set - it suddenly sounds like most of the stops have been pushed in with only a few still sounding as you play more notes. The virtual keyboards do still show the keys being pressed.

That just means that the registration/piece you're playing is exceeding the polyphony limit setting, and Hauptwerk is preventing new notes from sounding. To avoid that you would need to set the polyphony limit higher, which probably means you'll need to add more CPU power (given that you mentioned you get audio glitches if you increase the polyphony limit).

Hauptwerk's polyphony management system first attempts to fade out releases proactively and rapidly when the polyphony demand is nearing the limit. However, it takes time to fade out a release (an audible click would be heard due to the discontinuity if it didn't fade them). It will only prevent new pipes from sounding as a very last resort, i.e. if the polyphony limit has actually been reached completely and all sounding releases are still in the process of being faded out.

Whilst that situation is very unlikely to happen with medium registrations (because there will normally be plenty of releases from earlier-played notes that can be faded out safely in good time), it could happen with extreme registrations when hundreds or thousands of pipes are actively playing at a given instant.

In summary, I'm afraid you probably do just need more CPU power so that you can increase the polyphony limit accordingly.

Very large, very wet surround-sound sample sets do just need a huge amount of CPU power.

[mdyde]

P.S. If upgrading the computer/CPUs isn't a desirable option, you could try turning off some of the realism options for the sample set on 'Organ settings | General options':



E.g. disabling interpolation and the harmonic-shaping filters will allow you to set the polyphony limit much higher for the sample set, at the expense of some realism (especially tremulants and the wind supply model will be less realistic) and at the expense of temperaments not working (since they need interpolation to be on).

Whether it's better to use the surround-sound version of the sample set with realism features disabled, or just to use the stereo version with realism features enabled is probably a matter of personal preference.

Thanks Martin

The buffering issue is that I presume this (in crude terms) implies that the audio stream is fed into a memory buffer and comes out the other end after a short latent period. The buffer size can be adjusted with linked implications on CPU load and latency. Where does that memory reside?? Presumable in the RAM as used by the rest of the system. If there is an issue with memory handling / paging elsewhere then even a very small glitch in terms of memory access could slow down buffer access under higher stress CPU load or allow the buffer to empty and may significantly impact on the final audible output. I would presume this could manifest as the dropped pipes I am currently hearing?

I'm off home now and will post some more later after some tweaking

[mdyde]

Hello Iain,

The buffering issue is that I presume this (in crude terms) implies that the audio stream is fed into a memory buffer and comes out the other end after a short latent period. The buffer size can be adjusted with linked implications on CPU load and latency. Where does that memory reside?? Presumable in the RAM as used by the rest of the system. If there is an issue with memory handling / paging elsewhere then even a very small glitch in terms of memory access could slow down buffer access under higher stress CPU load or allow the buffer to empty and may significantly impact on the final audible output. I would presume this could manifest as the dropped pipes I am currently hearing?

The audio output buffers are allocated by the driver (probably within the kernel's memory space) and so should never get paged out. If they did, you'd get a severe audio glitch/crack, rather than some pipes not sounding.

You might conceivably get some pipes not sounding, or sounding late, if the data for some pipes had been paged out.

P.S. Chris:

Not sure if it is processor or memory related, as reducing the memory footprint (going mono, single loops etc.) does improve, but not remove the problem. I have 8GB RAM, and with the surround set loaded, I pretty much fill it.

Playing samples in mono puts about 30 percent less demand on CPU/polyphony than stereo, which probably accounts for that.

[micdev]

Dear all,

I think that we are dealing with two sepearate issues here.

- Memory swapping on the MAC
- High polyphony and CPU overload (Mac and PC)

Since I'm no pro with the MAC, I will leave this subject to those who know and understand this issue.

About the CAEN, while I'm still waiting the postman to bring me this set, I can report that I had a similar experience with the Metz (when using XP32 on a Pc with 3GB and a Core duo proc) and with the PAB Pro (Quad, 12 GB).

The main problem with big sample sets is that they are not only memory hungry, but can easily overload the CPU; once the cpu is overload Hauptwerk can't apply its algorithm to fade release tail; my experience with Gigastudio, Hauptwerk and other audio software shows me that we can get audio problems when the CPU reach a value as low as 45-50%, so the fact that the CPU doesn’t reach 80-90% doesn’t mean we won’t have problem.

With my old pc (xp32-3GB-Core Duo) the Metz cause the reported problem; moving to a Quad solved the problem, it wasn’t a memory related problem, but simply a CPU overload problem.

With the PAB Pro I sometimes have the problem when playing Tutti with full hands playing a fast passage; once again CPU overload. Don’t forget that the CAEN surround is a 110 ranks organ!

If you tick the General setting | General options | Audio Engine | Optimize for polyphony instead of model accuracy you can get 20-30% more polyphony (less overload)…. Some other options in that screen can also imprve the polyphony. Maybe less accurate, but it will be a lot more pleasing than all these drop notes and clicking. (Don’t forget to untick this option when switching organ, especially Theatre organ sets!)

Brett/Martin, would it be possible to have this option on an organ basis instead of a general option set-up; that way we could retain accuracy for those less resources hungry organs and use the less accurate modeling for the big one.

Best regards
François

EDIT: Oups, between the time I read the thread, wrote my text and post it, Martin provided more info... sorry for the dupicate info.

you might find you really want an 8-core computer to play its tutti

Thanks Martin for your detailed response(s). I suspected as much, now, do I have to tell the wife, or can someone else do it ;)

[mdyde]

Hello Francois,

If you tick the General setting | General options | Audio Engine | Optimize for polyphony instead of model accuracy you can get 20-30% more polyphony (less overload)…. Some other options in that screen can also imprve the polyphony. Maybe less accurate, but it will be a lot more pleasing than all these drop notes and clicking. (Don’t forget to untick this option when switching organ, especially Theatre organ sets!)

Brett/Martin, would it be possible to have this option on an organ basis instead of a general option set-up; that way we could retain accuracy for those less resources hungry organs and use the less accurate modeling for the big one.

Yes - we do plan to move that particular setting to the organ-specific options screen in the near future.

Ok Guys - update - fast toccata style playing used

Caen surround loading in 9.6 GB with 40 stops (ie equiv of 80) pulled generates 30% CPU load, polyphony set@5500- unplayable after 4 - 6 bars. Tweaking all the boxes re harmonic filters etc - No difference.

(First encountered on 50 stops and all Oct Grave couplers trying to push system got up to 50% CPU and unplayable after 2 bars)

Reloaded at 7.7GB (Around 8.5GB for a sample set is the effective "2/3" limit on my machine). Sounds significantly less good, but playable at 40 stops giving 24%CPU load . 50 stops + OG coupler still drops pipes @ 30% CPU load - Not unplayable, but un fulfilling to play!

Lets compare to Metz extended - Loads at 9.4 GB - above the advised safe limit. All 48 stops with all grave and super octave couplers pushes the CPU to 54% with a few drop outs audible to the trained ear - still very playable. I reckon this generates greater polyphony that Caen at 40 (ie 80 surround) with no octave couplers.

Sauer is similar.
(This is way above a registration I would choose to use in real life on these organs)

So.... why can I hammer one organ up to 54% CPU load with more than the equivalent number of notes playing simultaneously and still find it usable and another is unplayable at 30% CPU load ???

Are we really saying that if you want to play the music that the Caen Surround organ is designed to play you need an 8 core machine with 24GB?? Perhaps Chris has a point about length of reverb tails impacting on polyphony.

I suppose I should email Jiri and ask him to comment as to me this seems to be as much an issue with the sample set as the HW platform

I have one last idea to try - max out the memory with uncompressed samples to see if the reduction on CPU load helps. I need a G&T first though :-)

[mdyde]

Hello Iain,

Caen surround loading in 9.6 GB with 40 stops (ie equiv of 80) pulled generates 30% CPU load, polyphony set@5500- unplayable after 4 - 6 bars. Tweaking all the boxes re harmonic filters etc - No difference.

Did you also increase the polyphony limit? If not, it wouldn't make a difference.

E.g. on 'Organ settings | General options' try:

- Disable harmonic-shaping filters.
- Disable swell box filters.
- Disable per-pipe EQ.
- Disable interpolation.
- Polyphony limit = 12000.

.. with an audio buffer size of 1024 for all outputs on 'General settings | Audio outputs'.

Did you also increase the polyphony limit? If not, it wouldn't make a difference.

Doh!!!! No - must have been having a Homer Simpson moment - will give it another go later.

[micdev]

Ok, I've received the Caen, installed the 8 DVDs, loaded 16-bit (full releases, loops, etc etc)

Intel Quad 9450
16 GB.
Load in 11.2 GB
Polyphony - 5500, then 6500 and finally 7500.

Result: CPU overload! click, clanks, pouf!! Normal, with a 110 ranks organ.

The effect/problem is exactly the same I had with the Metz and my Core Duo; the solution was to upgrade to a Quad Core. As I already said, it is not a memory issue, just a lot of ranks and the CPU working overtime.

You want to play Tutti, surround, fast full chords pieces.... better get a big CPU. But in the meantime, don't be sad, you still can play the wet version or surround only version redirected to the front and eventually (depending how wet you like it).

Take advantage of the offer and spend another 25E even if your computer can't take it, eventually it will... Also, on a less "tutti registration", thanks to the surround you can get a pretty full sound; so even if you can't use the Tutti, you can have plenty of sound using the surround and a lesser registration.

I just love it.... to get up in the morning I need motivation to go to work and earn some money... I now have a new motivation... money for a newer, faster, better computer for the PAB and Caen of this world (and don't forget the upcoming volume 2-3 of the Bovenkerk...)

Have fun ;-)

François

Also, on a less "tutti registration", thanks to the surround you can get a pretty full sound; so even if you can't use the Tutti, you can have plenty of sound using the surround and a lesser registration.

I agree, in my opinion too, it's not necessary to select all stops for a nice 'full organ' sound.
For a real organ you need all stops to create sound volume, for Hauptwerk you can create volume with your amplifier.
Again, also for 'colloring' not all stops are needed.

BTW My opinion about Caen is on: http://www.pcorgan.com/SampleSets1885CaenEN.html

All the Best,
Gert

OK - finally got somewhere!!

I will give you my theory/reasoning!

1. Polyphony is the key. I need a number of 8500 to make enough of the organ work to keep me happy. With all the realism features disabled and a modest load of 7GB I pushed up the polyphony till the organ spoke reliably. I then put realism features back on one by one until the pops and cracks appeared. (Still not sure why polyphony maxes out and fails at modest CPU load, but it does - have to accept that as something I can't change)

2. Memory is not an issue - ie fill the mac memory to 80%, all it does is slow down the load a little. It has no bearing on performance in itself - but -- I did get an improvement when I reduced the size of the cache. I Think this was because I went from 20 bit all loops to 16 bit single loops - the CPU load to decompress the smaller files must be less - therefore polyphony improved. I have now filled the rest of my memory with uncompressed stops (the stops always used) up to 9.6GB and this has allowed me more realism features.

I have lost interpolation (and therefore temperaments) and per pipe EQ. I can live with this for now. Having selected these on the Organ Settings rather than General Settings, I still have these enabled on all my other organs. (I have checked - I can still load historic temperaments on the others, not Caen)

Not sure if I can set General Settings Polyphony to 0 and specify polyphony for each organ individually in the organ settings - will try that tomorrow.

I don't like the wet set alone - too dry with the upper work sounding even dryer. The surround for me is well worth the lack of interpolation.

It really is a juggling act to balance all the inter-related parameters. Having never had an issue with polyphony before this was a bit a voyage of discovery.

For info I found the most taxing piece for polyphony was the Boellman Gothic Toccata played v fast. I knew that one day I would find a use for it :-)

Many thanks to all for their comments.

[B. Milan]

Not sure if I can set General Settings Polyphony to 0 and specify polyphony for each organ individually in the organ settings - will try that tomorrow.

Hi Iain,

No, but you can override the General polyphony settings by specifying something other than 0 in the organ settings polyphony settings. So, you can still set the polyphony for each organ individually by just doing so in the organ settings portion.

Simply put, if the organ settings polyphony are set at 0 then Hauptwerk looks to the general settings polyphony, if it is something other than 0 then it just uses that.

Could these problems be due to the bandwidth between the CPUs and RAM? Since the CPUs are not maxed out, there must be some other bottleneck in the system.