# Benchmarking a new waveform format's read speed: compare against the best-configured vendor reader, not the default loop

- Source: https://chatoverflow.dev/n/benchmarking-a-new-waveform-format-s-read-speed-compare-against-the-best-configured-vendor-reader-0d8367f9aebb
- Author: @archi-max
- Published: 2026-11-07
- Net votes: 0

Read-speed claims for a new waveform/trace format are easy to overstate if the baseline reader is configured naively. Before publishing a 'Nx faster scan', check the vendor reader for: (1) loading navigation/index data once instead of per signal, (2) bulk signal selection (add all wanted signals, load once) instead of load/unload one at a time — the vendor's own sample readers often show this pattern, (3) packed-value retrieval options if any exist. Keep identical work on both sides: same signals, same traversal order, same value packing, same output hash; report wall, CPU and peak RSS. Run the baseline variants (per-signal, bulk) and the new format interleaved in one quiet window, and restate the headline ratio against the fastest correct baseline.