Version 3.5 much more demanding on CPU

Yup, I confirm. Beta version of 3.5 never clicked or chopped the sound for me, but 3.5 final gives me constant crackling on fairly large latency (512 samples). That is not satisfactory.

I'm not seeing any performance regression on my system -- the PI seems to average at 37, and I'm running the program at 48kHz and 128 samples for 256 voices.

I did find that ASIO4ALL gives me crackels and pops once the sample buffer gets bellow 384. This wasn't an issue in version 3.05
But I also found that if I check the "Use Hardware buffer" in ASIO4ALL - they go away.
Also - my native MAudio Profire 610 ASIO drivers seem to behave nicely... so maybe it has to do with software vs. hardware buffer.... ???

etalmor - I tried activating hardware buffer and it surely didn't have THAT MUCH clicks when over 100 poly. But it still choked on 100 poly part in my MIDI.

http://www.forum-pianoteq.com/uploads.p...hunder.mid

Try this file, people, please, and tell me your experiences!

Hahaha Sorry, I should've given the heads up ^^'

BTW; is there any chance for you guys to try playing this with ASIO4ALL driver? I suspect the problem is there, and am talking to Julien about it over e-mail as we... post

My A4A settings:

512 samples (if I go below, hell awaits), latency compensation 32/32 samples, hardware buffer on (0 ms), resample 44.1->48 off, force WDM to 16-bit off.

Realted HD onboard soundcard -_-'

Hmm, yeah it is of course always preferable to use your soundcard manufacturer's own native ASIO drivers, if they have any.
Asio4All is a last resort.

I use a Core2Duo with 1.67 Ghz. I can play Pianoteq 3.5 with 192 samples, 48kHz and ASIO4ALL with hardware buffers switched off (built in Realtek sound chip). With 128 samples I hear cracklings.

Because this sound chip is crap, I normally use an Audio Kontrol 1 with native ASIO driver, again with 192 samples and 48kHz. Less samples are not possible here as well.

In both setups I have to limit the polyphony to 96. Otherwise I get CPU overloads. And I cannot see any difference to Pianoteq 3.0.5. I did not change the settings after upgrade to 3.5.

My problem ist that the Audio Kontrol driver does not work with Windows Vista or Windows 7 on my hardware. I have to stay with Windows XP.

No, the CPU load is FINE. The drivers are what is causing the clicks.

Don't buy Asus soundcards -_-' (Their motherboards aren't very good lately either.) Get external USB ones, like E-MU 0404 USB for example. Those have very good low latency drivers.

Also, AMD "equivalent" is not really 11.2 GHz single core That's just a marketing gimmick, don't buy it.

Thanks for the info - the issue really seems to be A4A.

I'm using Pianoteq 3.5 on a Linux Fedora/PlanetCCRMA system using PlanetCCRMA's rt-kernel and jack audio system.

My Pentium IV laptop is a little bit outdated to get zero x-run performance.
I can't get crackle-free sound if I want at the same time the smallest latency and vice versa. I want a new computer one day, but I'll wait for Intel i7/i5 technology to be available and affordable on laptops.

For the time being I can live with a little bit of crackling in my sound.

Which brings me to the following question: how good do linux systems (Fedora/PlanetCCRMA, Ubuntu Studio ?) , with rt-kernel and jack audio system, perform compared to Windows systems and drivers running on the same or similar hardware and processors ? I can't compare, since I don't have a Windows box.

I too have more problems with crackling than with the previous version, but the main reason in my case is the ASIO driver of my Tascam US-122 that takes 12-13% cpu just looping idle on my AMD dual core. If the driver and pianoteq run on the same core, the driver might get interrupted more often, resulting in audio crackles.

In previous versions, it seems I had somehow a better chance of the driver starting on a different core than pianoteq itself, resulting in a better balanced system, but with 3.5, most of the time they are on the same core, maybe due to a different order of thread launching by pianoteq.

I tried running pianoteq alone or starting the Tascam user interface first but nothing is working all the time.

Now, I know AMD is much worse than Intel in balancing threads even if I already run the optimisation furnished by AMD, but I wonder if it would not be possible somehow to FORCE the ASIO driver to start on a different core when available.

The driver and pianoteq are in a client-server situation and I think this would improve things for everybody anyway.

I don't have this problem with my other card (Audigy 2ZS) since everything seems done in hardware, no cpu usage is visible.

I haven't done any multi-thread programming, and maybe this is not possible in a general way and has to be left to the destination hardware to decide, but I would appreciate if Julien could shed some light into this.

Thanks for the advice, but you are right, it is worse when locking to one cpu. The problem is there is only one pianoteq process running and the loop that takes too much resource in my case is probably a thread inside the process, part of the Tascam ASIO driver.

Sometimes this thread gets launched on the other cpu and that helps very much, but I have no control on the thread assignement.

Of course this might be because the Tascam driver is badly written and shouldn't have such a loop, but I can't update it since Tascam doesn't sell the US-122 any more, and ASIO4ALL doesn't seem to be a very good choice.

If the ASIO driver was launched as a different process, then I would be able to lock it after launch but there would probably be a performance penalty with interprocess communication.

You can have both 3.5 and 3.0.5. installed in different directories. Some DAWs won't see both VST DLLs, though.

I looked into it, and the US-122L and US-144 use a completely different driver numbering scheme than the obsolete US-122, so I guess they would not be compatible since the newer models support 96kHz samples while the US-122 is limited to 48kHz. Even these models are now obsolete, having been replaced by a "MKII" version.

The Tascam drivers are annoyingly complicated to install and I wouldn't try it unless I have a really good reason...

I did find (by googling) that a newer driver (3.4.0) exist for the US-122 but it reportedly only adds Vista support which I don't need.

Tascam is not very reputable for software support, having completely stopped selling GigaStudio, which was the reason I had bought the US-122 in the first place...I have to live with my choice for the time being since I find that the 24 bit US-122 sounds very good with pianoteq, better in fact than the other card I have, the Audigy 2ZS.

Windows XP SP2, Intel Core2Duo 2.0 GHz, 2 GB DDR2 RAM. ASIO4ALL is the issue.

3.5.1 Pro behaves much better now.

I think I have found a workaround for my dual core balance problem but I find it a bit puzzling.

I first launch the US-122 control panel which starts running under a standard process known as run32dll.exe, then I use its guitar tuning function which puts it in a loop waiting for guitar input, then I set the processor affinity of the run32dll.exe process to cpu 0 EVEN if it already runs on cpu 1.

After that, I launch pianoteq and it starts running on cpu 0 and I get my balanced system where I have both cores running equally, and I get no crackles. I then have to quit the US-122 control panel with the ok button, or it will go on waiting for guitar input forever...

This doesn't work if I don't lock rundll32.exe or try to lock it to cpu 1 where it is already running, pianoteq then, most of the times, starts also on cpu 1...the other running tasks don't seem to have any effect on this behaviour.

The only explanation I can think of is that, on my hardware, Windows mostly uses cpu 0 and the system has a tendency to start cpu intensive processes on cpu 1 first. By starting first the driver loop, the system then uses cpu 0 as a first choice for pianoteq which comes later. But why it work only by setting this false affinity for the rundll32.exe is beyond me! The driver loop STAYS on cpu 1 it doesn't move to cpu 0.

theJourney:

Have you tried the usual experiments with increasing the number of buffers for the sound card?

Something else you might try: A few months ago, someone posted that he was getting surprisingly good sound with the PTeq sample rate reduced to 2200. You'll notice some loss of quality, but not as much as you might expect,and it might be viable for playing in the hotel room late at night.

The general rule is to increase the buffers until the latency is unaccepatable. (The buffers hold data\samples into short term memory, so they can be played rapidly). But at some point, increasing them will start to cause too great a latency, since the system has to load the samples into them. There's no rule about the correct number--you have to experiment. The goal is to find the least number possible. Just start with very high settings, listen to the results (high latency) and then reduce them stepwise until you reach the point at which you get break-ups, and then back-up one step.

But if you have a problem with other sound software not producing sounds while you play Pianoteq, the problem is probably that the internal card just won't let you use both sound sources at once. Can you play two other sound-based programs at once? That shouldn't happen. Assuming your on a Windows machine, did you open the Mixing panel? Something there may be muted or reduced in volume.

Regardless, you may want to invest in one of the several relatively inexpensive external usb or firewire sound cards. They start at around $150, fit in a briefcase or satchel, and produce much better and much louder (2-3X) the sound.

But think about what specs you want. Most have at least one input and output jacks, so that, when you're at home, you could plug in other several other instruments as well and\or run lines to powered monitors. The more inputs, the better. Some have mics inputs. Most, to use a mic for a vocal or recording an acoustic guitar, etc, require that you use an additional amplified mixing board. (These are also readily available, small, and inexpensive, and you could get one later, after you get the basic setup.)