I put together a PREEMPT_RT Linux box (PianoRT) to run Pianoteq at very low latency, and I'm curious how it compares to other setups.
It boots straight to Pianoteq in under 5 seconds, holds 1.3ms measured buffer latency, and handles 250+ voices of polyphony with zero xruns, even with the sustain pedal held down through a full piece.
Video of it running: https://youtu.be/X46wiM3ON64
If you want to try it on your own machine, the benchmark file and rules are here: pianort.com/index.php/the-250-challenge/
Curious what people find.
Really great idea, project and optimization process.
I tried the test on my Linux machine, an old laptop, and tried as many optimizations as I could think of: CPU-power GUI to get rid of frequency scaling (set all cores to maximum frequency continuously), downloaded and installed the linux-lowlatency kernel as well as the Liquorix kernel, added preempt=full and threadirqs to the boot parameters, and all of that helped tremendously.
But when I tried to play the test MIDI file in Pianoteq running with 64-sample buffer at 48000 sample frequency, and 256 voice polyphony, with the sustain pedal engaged continuously (with MIDI 64 disabled in Pianoteq's MIDI-mapping panel so that the sustain pedal recorded into the test MIDI file wouldn't trigger the sustain pedal in Pianoteq), and then opened two terminal windows to simultaneously run pw-top and cyclictest as they continuously (at regular intervals) updated the information in the terminal, while recording all of that in OBS Studio, including the desktop GUI and the audio output of Pianoteq, it was all a bit too much for this 6-7 year old laptop at several points during the playback when the maximum polyphony was reached, and there were between 1-3 fairly brief audio dropouts with all of that going on, in several consecutive attempts.
My system specs--
CPU: 6-core Intel Core i7-9750H (-MT MCP-) speed/min/max: 4366/800/4500 MHz
Kernel: 7.0.0-34-generic x86_64 Up: 1h 5m Mem: 7.37/31.05 GiB (23.7%)
Storage: 2.27 TiB (46.7% used) Procs: 450 Shell: Bash inxi: 3.3.34
Drives:
Local Storage: total: 2.27 TiB used: 1.06 TiB (46.7%)
ID-1: /dev/nvme0n1 vendor: Samsung model: SSD 970 EVO Plus 250GB
size: 232.89 GiB
ID-2: /dev/nvme1n1 vendor: Samsung model: SSD 970 EVO Plus 250GB
size: 232.89 GiB
ID-3: /dev/sda vendor: Samsung model: PSSD T7 Shield size: 1.82 TiB
type: USB
BOOT_IMAGE=/vmlinuz-7.0.0-34-generic root=/dev/mapper/vgmint-root ro quiet splash preempt=full threadirqsDon't you think the result vary with the preset used? There are preset that are heavier on cpu than other.
You should maybe specify a preset that is free from the kivir set, everyone have them.
Yeah, that's a great point. I only have the Kawai, and Steinway packages. I'm definitely open to suggestions.
]]>PianoRT wrote:I put together a PREEMPT_RT Linux box (PianoRT) to run Pianoteq at very low latency, and I'm curious how it compares to other setups.
It boots straight to Pianoteq in under 5 seconds, holds 1.3ms measured buffer latency, and handles 250+ voices of polyphony with zero xruns, even with the sustain pedal held down through a full piece.
Video of it running: https://youtu.be/X46wiM3ON64
If you want to try it on your own machine, the benchmark file and rules are here: pianort.com/index.php/the-250-challenge/
Curious what people find.
This is pretty cool. But I don't want to run it because I want to live in the belief that my rig is enough, haha.
Thank you for that! I've been working on this project since last December, and I finally am willing to share. I'm really looking for a Deathmatch competitor to see how it holds up outside of my personal testing and for a fun playful challenge. Theoretically PianoRT should work for Pianoteq and Organteq. The custom preempt-rt tuned for the Dell 3050 micro with coreboot *should* be able to maintain those 250 simultaneous voices, but I really like some real-world testing. Again, thank you for the compliment, it was a personal project that kept growing (maybe a little out of control) :]
Nick
]]>I put together a PREEMPT_RT Linux box (PianoRT) to run Pianoteq at very low latency, and I'm curious how it compares to other setups.
It boots straight to Pianoteq in under 5 seconds, holds 1.3ms measured buffer latency, and handles 250+ voices of polyphony with zero xruns, even with the sustain pedal held down through a full piece.
Video of it running: https://youtu.be/X46wiM3ON64
If you want to try it on your own machine, the benchmark file and rules are here: pianort.com/index.php/the-250-challenge/
Curious what people find.
This is pretty cool. But I don't want to run it because I want to live in the belief that my rig is enough, haha.
]]>It boots straight to Pianoteq in under 5 seconds, holds 1.3ms measured buffer latency, and handles 250+ voices of polyphony with zero xruns, even with the sustain pedal held down through a full piece.
Video of it running: https://youtu.be/X46wiM3ON64
If you want to try it on your own machine, the benchmark file and rules are here: pianort.com/index.php/the-250-challenge/
Curious what people find.
]]>