Linux Audio Benchmarks: Tests and Archives
Pick a benchmark that answers the problem you can reproduce. Timer wake-up delay, graph errors and latency measured through an interface are three separate results. A storage test answers a fourth question: can the disk keep up with recording and playback?
Choose a measurement
- Latency measurement: choose between scheduler testing, PipeWire diagnostics and a physical loopback.
- USB interface test method: keep the physical path and settings comparable, and record the spread across repeats.
- Audio quality: investigate the signal path, noise and conversion separately from timing.
Read a trace or historical result
The Latency Graph archive explains how to plot a time series and what a summary cannot tell you. The HDRBench archive documents Benno Senoner’s 2000 multitrack hard-disk benchmark. It did not measure scheduler latency. The local files listed under the old download names are not verified runnable copies.
Compare one change at a time
Save the baseline and record versions, test duration, device, workload and observed settings. Change one factor and repeat. Look at worst cases and run-to-run variation as well as typical behaviour. If the difference is smaller than the variation across equivalent runs, record the result as inconclusive rather than picking a winner.
Include both a simple baseline and a run that reflects the real session. For a disk question, use the same project, track count and sample format and record whether dropouts coincide with storage activity. For a graph question, replay the same passage under the same rate and buffer and compare the relevant counters. For noise or distortion, keep gain and the signal path fixed instead of turning a timing test into an audio quality score.
Record the first sign of failure as well as the headline number. A disappearing interface points to a different investigation from a plugin repeatedly missing the graph deadline. Go back to the practical setup checklist if the routing or device is not stable enough to repeat a test. After a change, return to your original settings for a control run if you cannot otherwise account for a difference.