Page 2 of 2

Re: Nancy -- some polyphony tests

PostPosted: Sat Jan 22, 2022 3:29 pm
by mnailor
Thanks for that -- I don't feel quite so stupid now.

One bad thing about running with Speedstep and Turboboost enabled is that the CPU speed stays elevated even when the system has no apps running, so I'll have to turn it off when not playing to avoid wear from heating and running the fan more. Speedstep doesn't seem to have any effect when Turboboost is one, unless Task Manager is the load that's keeping it from idling.

It also doesn't turn on the monitor if I allow it to sleep and then wake it up, so sleep on idle timer isn't a solution.

FIXED: Changed to Balanced power scheme from Custom, then reapplied the settings exactly as I had them in Custom, fixed it so the CPU speed goes the full range.

Re: Nancy -- some polyphony tests

PostPosted: Sun Jan 23, 2022 10:15 am
by mnailor
Here are some polyphony tests with the i9-9900 *not* locked at its base speed, 3.1 GHz. Speedstep and Turboboost both have to be on because Dell... During the tests, the clock speed ran in the 4.2 to 4.7 GHz range.

15 note chords, 14 manual and 1 pedal, all coupled, repeated at increasing speeds up to 4 chords per second. 90 seconds per test. Used MIDI playback for consistent tests. Amplifiers off. Varied the polyphony limit to find the highest limit that kept the CPU meter green, then yellow, but never red.

Hauptwerk 6 Higher def pitch shifting enabled throughout.

The two outcome numbers are the highest polyphony limit in k (multiples of 1024) that kept the Hauptwerk CPU meter in the green and yellow ranges, respectively. Any setting in this range should be safe from CPU-related audio glitches on this processor.

Format: sample_rate_khz/buffer_size_samples ___ max_green_polyphony_k_voices - max_yellow_polyphony_k_voices

Caen 2.66, 2 perspectives, all samples, full releases, 24 bit:

96k/0512 ___ 2.5 - 4.0k
96k/1024 ___ 3.0 - 5.0k
48k/0256 ___ 5.0 - 8.0k **
48k/0512 ___ 6.0 - 9.0k **

Nancy 1.02, 4 perspectives, all samples, full releases, 16 bit to fit in 64 GB memory:

96k/0512 ___ 1.5 - 2.0k
96k/1024 ___ 2.5 - 3.0k
48k/0256 ___ 3.0 - 4.5k
48k/0512 ___ 4.0 - 7.0k **

** The i9-9900 is adequate for the modes marked with ** with HW 6 higher def pitch shifting, without much polyphony loss. Since these are stress tests at full organ, you may be able to exceed these "safe" polyphony limits, depending on the music you play.

Note: Every Intel Core i9 generation 11 (i9-11xxx) and i9, i7, or i5 generation 12 (i*-12xxx, not i3) processor is faster than this one.

Re: Nancy -- some polyphony tests + Hauptwerk 7

PostPosted: Wed Jan 26, 2022 11:03 am
by mnailor
Update for Hauptwerk 7 on the i9-9900, same test conditions and method as above, except:

Both of the Hauptwerk 7 "higher" audio engine options were enabled.

The two outcome numbers are the highest polyphony limit in k (multiples of 1024) that kept the Hauptwerk CPU meter in the green and yellow ranges, respectively. Any setting in this range should be safe from CPU-related audio glitches on this processor.

Format: sample_rate_khz/buffer_size_samples ___ max_green_polyphony_k_voices - max_yellow_polyphony_k_voices

i9-9900 -- Caen 2.66, 2 perspectives, all samples, full releases, 24 bit, HW 7 best audio quality:

96k/0512 ___ 2.0 - 4.0k
96k/1024 ___ 3.0 - 4.5k
48k/0256 ___ 4.5 - 8.0k
48k/0512 ___ 6.0 - 9.0k

i9-9900 -- Nancy 1.02, 4 perspectives, all samples, full releases, 16 bit to fit in 64 GB memory, HW 7 best audio quality:

96k/0512 ___ 2.5 - 3.0k
96k/1024 ___ 1.5 - 3.0k
48k/0256 ___ 4.0 - 5.0k
48k/0512 ___ 4.0 - 7.0k

Re: Nancy -- some polyphony tests -- on the new i9-12900K

PostPosted: Thu Jan 27, 2022 7:51 pm
by mnailor
Here are some polyphony tests with the new Dell XPS 8950 with i9-12900K and 128 GB after a long day of Windows 11 performance tuning. SpeedStep, all cores, and HyperThreads are enabled in BIOS, but SpeedShift is *disabled* since it made Hauptwerk performance too unstable to measure reliably. Note that this BIOS doesn't let you turn off TurboBoost, but that turned out just fine.

I set the High Performance power plan with nothing allowed to sleep, processor min - max at 100 - 100%. Under load, CPU speed goes up to 4.8 GHz at times, closer to 4.3 when a lot of cores are busy. It steps down to about 2.6 GHz when idle, a bit lower than the 2.8 average of base speeds on 8 performance (P) cores at 3.2 and 8 efficiency (E) cores at 2.4, so the 100% minimum is not strictly obeyed. But stepping down keeps the fans from getting loud.

Tested with 15 note chords, 14 manual and 1 pedal, all coupled, repeated at increasing speeds up to 4 chords per second. 90 seconds per test. Used MIDI playback for consistent tests. Amplifiers off. Varied the polyphony limit to find the highest limit that kept the CPU meter green, then yellow, but never red.


The two outcome numbers are the highest polyphony limit in k (multiples of 1024) that kept the Hauptwerk CPU meter in the green and yellow ranges, respectively. Any setting in this range should be safe from polyphony-related audio glitches on this processor.

Format: sample_rate_khz/buffer_size_samples ___ max_green_polyphony_k_voices - max_yellow_polyphony_k_voices

1. Hauptwerk 7 audio engine set to higher audio quality and higher trem/wind/relay/etc models.

i9-12900K -- Caen 2.66, 2 perspectives, all samples, full releases, 24 bit, HW 7 best audio quality:

96k/0512 ___ 8 - 12k
96k/1024 ___ 9 - 12k
48k/0256 ___ 15 - 20k
48k/0512 ___ 18 - 30k

i9-12900K -- Nancy 1.02, 4 perspectives, all samples, full releases, 24 bit, HW 7 best audio quality:

96k/0512 ___ 5 - 8k
96k/1024 ___ 6 - 10k
48k/0256 ___ 11 - 15k
48k/0512 ___ 16 - 20k

2. Hauptwerk 7 audio engine set to medium audio quality and medium trem/wind/relay/etc models. Legacy HW 6 quality.

i9-12900K -- Caen 2.66, 2 perspectives, all samples, full releases, 24 bit, HW 6 best audio quality:

96k/0512 ___ 8 - 12k
96k/1024 ___ 9 - 13k
48k/0256 ___ 13 - 20k
48k/0512 ___ 20 - 31k

i9-12900K -- Nancy 1.02, 4 perspectives, all samples, full releases, 24 bit, HW 6 best audio quality:

96k/0512 ___ 6 - 9k
96k/1024 ___ 7 - 12k
48k/0256 ___ 10 - 15k
48k/0512 ___ 15 - 21k

Re: Nancy -- some polyphony tests

PostPosted: Fri Jan 28, 2022 11:59 am
by mdyde
Thanks, Mark.

That's a lot of polyphony!

[I've moved the topic here, since it mainly covers polyphony achievable on computer hardware.]

Re: Nancy -- some polyphony tests

PostPosted: Fri Jan 28, 2022 12:53 pm
by mnailor
Martin, thank you. I would have moved it there if I could.

Couldn't quite get Caen up to 32k polyphony without hitting one red bar on the CPU meter.

Re: Some polyphony tests

PostPosted: Tue Feb 01, 2022 11:50 am
by mnailor
I uploaded an Excel spreadsheet of polyphony tests to google drive and set it to share with anyone who has the link.

Added some tests on an AMD Ryzen 7 5700U laptop, but not Nancy in 16 GB, obviously. I'm using that from a single keyboard.

Not sure how well this will work, or what the download and view quotas for my free account are.

https://docs.google.com/spreadsheets/d/ ... ue&sd=true

Re: Some polyphony tests

PostPosted: Tue Feb 01, 2022 2:55 pm
by IainStinson
Thanks for this. The download (51k) works fine.
Iain

Re: Some polyphony tests

PostPosted: Fri Feb 11, 2022 9:43 am
by mnailor
I added 3 MIDI files to the google drive directory in case anyone wants to run their own tests. Also see Methodology tab in spreadsheet.

Here's the link to the directory containing the updated spreadsheet and MIDI (.mid) files:

https://drive.google.com/drive/folders/ ... sp=sharing