Linux Audio Benchmarks

Measurement is one of the areas where Linux audio work has historically been both rigorous and poorly communicated. The people doing serious latency measurement on Linux have produced genuinely useful data, but that data has been scattered across mailing list threads, personal sites, and presentation materials from community events. This page is the hub for all the benchmark tools, latency traces, and measurement resources available on this site. I cover what each resource contains, how to use the tools, and how the different measurement resources relate to each other for anyone trying to build a complete picture of audio system performance on Linux.

Why Measurement Matters in Linux Audio

Audio work on Linux involves performance requirements that most software does not face. The latency between a sound event and its reproduction must be low and, more importantly, predictable. A system that delivers low average latency with occasional spikes is not acceptable for real-time audio production. The spike is audible. Getting a Linux system to deliver consistent low latency requires understanding at several levels: the audio hardware and its driver behavior, the kernel scheduling configuration, the buffer sizes chosen in JACK or ALSA, and the load the system is under during operation.

Benchmarking in this context is not about peak performance in the way that CPU or memory benchmarks are. It is about characterizing the worst-case behavior of the system, the tail of the latency distribution rather than the average. A system with a median latency of 5 milliseconds that spikes to 50 milliseconds twice a minute is worse for audio work than a system with a median latency of 8 milliseconds that never spikes at all. The tools on this site measure and visualize that kind of behavior.

HDR Benchmarks

The HDR benchmark tools available at hdrbench cover high-dynamic-range measurement approaches for audio system evaluation. HDR in this context refers to the range of the measurement itself rather than to visual display technology: capturing both very small and very large values with meaningful resolution across the full range. Standard histogram approaches for latency measurement can lose information in the tails of the distribution - exactly the part that matters most for audio work. HDR measurement techniques preserve that information.

The hdrbench tools are designed for users who want to characterize system behavior accurately rather than just establish a headline latency figure. If you are comparing two kernel configurations, two hardware setups, or two versions of a driver, you need measurement that captures how the configurations differ in their worst-case behavior, not just their typical behavior. The HDR approach makes that comparison possible and interpretable.

Latency Graphs

The latency-graph section provides visual tools for analyzing latency traces from running audio systems. Raw latency data from measurement runs is difficult to interpret in tabular form. The distribution of latency values, the pattern of spikes, and the relationship between system load and latency behavior are all much more legible in graphical form than as numbers. The latency graph tools on this site are aimed at making that visualization process straightforward for people who have collected latency traces and need to understand what they are looking at.

Visualization is particularly useful when you are comparing configurations or diagnosing a problem. A graph makes it immediately apparent whether a particular kernel configuration produces occasional large spikes or steady elevated baseline latency - two different problems that call for different solutions. The latency graph tools support both exploratory analysis and documentation of results for sharing with others.

Latency Measurement Reference

The latency section is the primary reference on this site for latency measurement methodology. It covers how to measure latency accurately, what tools to use, how to interpret results, and what the common measurement mistakes are. Latency measurement has a number of subtle failure modes: measuring the wrong thing, using tools that introduce their own latency into the measurement, or interpreting results in ways that miss the information relevant to audio production.

For Linux audio developers who want to understand the latency resource material that came out of the LAD community discussions, the resources page at resourceslatency covers the latency references that were discussed and maintained by the LAD mailing list community. These complement the measurement methodology material with community knowledge about specific hardware, kernel configurations, and real-world deployment patterns.

Audio Quality

Audio quality measurement extends beyond latency. The audio quality section covers the broader set of measurement concerns for Linux audio systems: signal quality, dynamic range, frequency response, and the various ways that software signal processing can introduce artifacts or degrade the audio path. These measurements matter for anyone doing recording or processing work where the quality of the signal chain is as important as the latency of the system.

Quality measurement and latency measurement address different aspects of audio system performance and are best used together. A system might have excellent latency characteristics but poor signal quality due to driver-level processing or hardware limitations. Conversely, a system with excellent signal quality might have latency behavior that makes it unsuitable for real-time performance work. The resources on this site cover both dimensions.

Using These Resources Together

The benchmark and measurement tools on this site are most useful when used in combination. Start with the latency reference at latency to understand what you are measuring and why. Use the tools at hdrbench to collect accurate high-dynamic-range measurements from your system. Visualize the results using the tools at latency-graph to understand the shape of the latency distribution. Cross-reference with the LAD community latency resources at resourceslatency to understand how your results compare to known hardware and configuration patterns. And use the audio quality section at audio quality to evaluate the signal path independently of the timing characteristics. Together these tools give you a complete view of your Linux audio system's performance.