Pianoteq Deathmatch: Dell 3050 vs Raspberry Pi, referee Nick

The PianoRT Benchmark

Think your Pianoteq setup can hold 250 voices clean? Prove it.

PianoRT holds 1.3ms buffer latency and 250+ voices of clean polyphony under heavy sustain pedal, zero xruns, over a 220 second stress run. 250 is the real number, not a round one, we start cracking slightly above it too.

1.3ms
buffer latency
250
voices, zero xruns
64
sample buffer
2
passes required

The Test

Two passes on the same file, buffer fixed at 64 samples (same as PianoRT), polyphony set to 250 voices. Sample rate is your choice, just report it.

Use the NY Steinway D Classical preset. It ships with every Pianoteq edition (including the free demo), so everyone is testing the same instrument.

Pass 1
As written

Load the file and play it start to finish exactly as written, pedal markings and all.

Pass 2
Pedal stress

Same file, same 250 voices, but hold the sustain pedal down for the entire piece regardless of what’s written, and watch for crackles or xruns.

Zero xruns on both passes is a pass. Any dropout and you’re out.

Download the benchmark file

Download the MIDI file

What to submit

To count, I need the real data, not “it sounded fine to me.” Three things:

Your rig
CPU, RAM, storage type, OS and version (kernel version if Linux), and the buffer size / sample rate you actually ran.
Real-time numbers
On Linux: sudo cyclictest -m -Sp95 -i200 -D 5m for max scheduling latency, plus a pw-top screenshot showing the Pianoteq and ALSA sink nodes, quantum/rate, and xrun count during both passes. On Mac/Windows, whatever latency or dropout monitor you’ve got, DAW-reported xrun counts are fine.
Proof
A screen recording or video of both passes, so it’s verifiable and not just self-reported.

Where this goes

Every real submission gets a shoutout. Best results (and the funniest failures) go into a PianoRT Machine Death Match video on the YouTube channel. Bring receipts.

“Omnis Lacrima” is from Final Fantasy XV Piano Collections, composed by Yoko Shimomura, used here for personal, non-commercial benchmark testing only.