Markdown to PDF Speed Comparison 2026 07 13

RayoMD 2.2.0 Three-Tool Windows Comparison

Run date: 2026-07-13.

Scope

This report compares fresh-process elapsed time for RayoMD, Pandoc/XeLaTeX, and Node md-to-pdf on three deterministic synthetic Markdown inputs. RayoMD was the fastest of the three tested converters in every case. This supports a scoped fastest-in-this-benchmark claim, not a universal ranking: the tools have materially different feature and rendering scope, and the result applies only to the recorded machine, versions, inputs, and options.

Results

Case	Input	RayoMD 2.2.0	Pandoc + XeLaTeX	md-to-pdf	Next comparator / RayoMD
Tiny README	205 B	15.45 ms (15.34-16.55)	3,210.69 ms (3,078.22-3,527.92)	1,717.89 ms (1,666.46-1,766.14)	111.2x
Medium feature mix	15,102 B	17.29 ms (16.49-18.03)	3,377.94 ms (3,355.08-3,641.19)	2,281.90 ms (2,261.89-2,330.59)	132.0x
500-row table	33,353 B	18.66 ms (16.97-19.63)	3,427.50 ms (3,383.43-3,575.74)	2,340.90 ms (2,305.30-2,408.92)	125.5x

Values are median milliseconds with min-max ranges across seven measured runs. The next-comparator ratio divides the lower comparator median by the RayoMD median; it does not compare rendering fidelity or feature completeness.

For context only, the 2026-07-11 RayoMD 2.0.0 medians on the same corpus and machine were 17.93, 19.74, and 30.64 ms. The new medians are 13.8%, 12.4%, and 39.1% lower respectively, but cross-day measurements are not treated as a controlled version-to-version benchmark.

Environment and Versions

• Gigabyte B550 AORUS ELITE system, AMD Ryzen 5 5600X (6 cores, 12 threads), 128 GiB RAM, Windows 11 Pro build 26200.
• RayoMD 2.2.0 native mode, Windows GCC 15.2 Release build.
• Pandoc 3.9.0.1 with MiKTeX-XeTeX 4.16.
• md-to-pdf 5.2.5 on Node 25.8.1 with Puppeteer/Chromium.

Method and Caveats

• Every measurement launched a fresh process and included startup, Markdown parsing, PDF generation, and output-file writing.
• One uncounted warm-up was followed by seven measured runs; the table reports the median and range. Each run overwrote the same destination.
• Inputs were deterministic local files with no remote resources.
• Every generated output was checked for a PDF header, EOF marker, and non-trivial size.
• RayoMD implements a focused native Markdown subset. Pandoc/XeLaTeX is a much broader publishing system, while md-to-pdf includes browser/CSS rendering and syntax highlighting. This is not a visual-equivalence comparison.
• Remote-image timing is excluded because network latency is not reproducible.

Reproduce

Install the Node comparator into ignored benchmark output if needed:

npm install --prefix benchmark-output\competitor-tooling md-to-pdf@5.2.5

Build RayoMD and run the maintained comparator:

cmake -S . -B build\windows -G "MinGW Makefiles" -DCMAKE_BUILD_TYPE=Release
cmake --build build\windows --config Release
python tools\benchmark.py competitors -- `
  --rayomd build\windows\rayomd.exe `
  --node-modules benchmark-output\competitor-tooling\node_modules `
  --root benchmark-output\competitors-2026-07-13 `
  --runs 7 `
  --timeout 180

The ignored output contains the generated corpus, PDFs, record.json with version and min/median/max data, and summary.md.