San Francisco Skinner CPU time

When loaded with no stops drawn, this organ has the Hauptwerk (8) process using 9 - 10% of the CPU steadily. No audio bus activity indicators are lit, so nothing is generating audio. I've disabled the blower and there's no convolution reverb or tremulant motor. Switching from sampled to modeled trems also doesn't change it.

I don't have any other organs with a noticeable background CPU utilization when idle [TURNED OUT NOT TO BE TRUE -- SEE BELOW]. Any ideas what this could be? Maybe the tremmed ranks are always running as a background layer?

Thanks!

[mdyde]

Hello Mark,

I have no experience with that organ myself, but as a test you could try disabling the wind model on the Organ Preferences screen, to see whether that's the culprit.

Good guess! Disabling the wind model dropped it from 10% to 5%. Then lowering model processing from highest to lowest dropped it to 2%.

Now that I understand it, I'm leaving it with wind model enabled and medium model processing for 9% steady CPU utilization. It doesn't keep me from using full organ.

San Francisco at 3 mic positions x 88 ranks is a fairly large organ.

Thanks!

[mdyde]

Thanks, Mark. You're very welcome.

Oops. Observation shortcoming. It seems I hadn't played some organs with Task Manager up to notice their CPU utilizations with the organs loaded and idle:

Doesburg: 10%
San Fran: 10%
Billerbeck: 8%
Nancy: 3%

[mdyde]

The background CPU usage is probably mainly due to the background models, which would be normal and probably doesn't indicate a problem. I recall from another thread that you run Hauptwerk at real-time priority and un-ticked the "Bind audio engine threads to CPU cores on Windows?" general preference. Conceivably you might get better performance if you ticked that preference again, since it should avoid the background models competing for CPU time with each other, and with the audio engine. (However, some people need to keep that preference un-ticked due to audio glitch problems from other things running on their PCs.)

Right, the background CPU use isn't a problem at all, I was just trying to identify it to make sure I hadn't configured something badly.

(CORRECTED BELOW) I do get measurably better max polyphony on *this* machine by letting Windows 11 place the threads, using realtime priority, and with turbo boost and hyperthreads enabled. No issues with that, even though it's not recommended. No audio glitches. There's no other CPU activity interfering with Hauptwerk, although Windows and Dell updaters keep re-enabling themselves somehow, the evil buggers.

I've changed sample rate/buffer from 96k/1024 to 48k/512 because 96k seems to bring out some harsh sounds in upperwork in certain samplesets, possibly because my audio interfaces do a better job at 48k. That means even more capacity available. I set max polyphony at 32k on all but two organs that I left at 16k, and they can take "forearm music" at full organ. Hauptwerk 8 really helped the i9-12900K.

Thanks!

[mdyde]

Thanks, Mark. Excellent.

I realized later that I hadn't retested systematically after HW7 using the same MIDI recording, so my assertion about binding threads worked for 7 but isn't true on 8. Binding and not binding threads gives equal polyphony results for me on HW8. The only difference is binding gives less even use of cores and runs at a higher clock rate and higher temperature -- so some more fan noise. On San Francisco only, the models keep one P-core at 100% all the time.

With sample rate/buffer 48k/512 and audio engine quality Higher/Higher, using full organ (no octaves graves) and 15 note chords, playing the same MIDI recording, the max polyphony setting that keeps the CPU meter green and the peak observed CPU utilization at that polyphony are:

Caen (2 mic positions) [Caen result is typical for most of my samplesets]
HW8 32k max polyphony, <= 80% utilization
HW7 18k max polyphony

San Francisco (3 mic positions)
HW8 18k max polyphony, <= 75% utilization
HW7 not tested

Nancy (4 mic positions)
HW8 17k max polyphony, <= 70% utilization
HW7 16k max polyphony

Binding or not binding threads gave the same numbers on HW8. Not binding threads in HW7 gave a little more polyphony, and that's the result above.

[mdyde]

Thanks, Mark.

Binding threads does have the significant benefit that the background models (wind model, etc.) may be able to perform better, because they won't be out-competed for CPU time by audio engine threads, so you might get better realism by binding threads.

I'll be doing that from now on. Thanks, Martin.