Test Plots
100 points
10,000 points
5 time series with legend
50 semi-transparent rects
200+ text labels
Faceted scatter + smooth
This vignette presents pre-computed benchmark results comparing
unigd rendering performance against base R graphics devices
and popular alternatives.
Benchmarks are generated offline by running
Rscript bench/run.R from the package root. The vignette
itself does not execute any benchmark code.
Each device renders every test plot using bench::mark()
with a minimum of 5 iterations. Output file sizes are captured from a
separate run.
Raster devices use 720 x 576 px; vector devices use 10 x 8 in (720 x 576 px equivalent at 72 dpi).
The timing tables show the median time and a ratio
comparing each alternative device to the corresponding base R device
(base::svg, base::png, etc.), so values below
1.0x mean faster than base R.
A note on Cairo backends: unigd uses Cairo under the
hood for PNG, PDF, and TIFF output. Comparisons with
Cairo::CairoPNG, Cairo::CairoPDF, etc.
therefore isolate device pipeline overhead (recording, serialization,
file I/O) rather than rasterizer differences.
The charts show individual iteration times as jittered points with a median crossbar, giving a sense of run-to-run variability.
Results were generated on Windows (x86-64, windows) with R version 4.5.2 (2025-10-31 ucrt) at 2026-02-15 23:34 UTC.
CPU: Intel(R) Core(TM) i7-10875H CPU @ 2.30GHz
Package versions: unigd 0.2.0, bench 1.1.4, svglite 2.2.2, ragg 1.5.0, Cairo 1.7.0, ggplot2 4.0.2
unigd’s deferred rendering architecture records graphics commands first, then serializes on demand. This avoids redundant intermediate state and tends to show an advantage on complex plots.
| Plot | base::svg | svglite | Cairo::CairoSVG | unigd svg |
|---|---|---|---|---|
| scatter_small | 6.2 ms | 6.4 ms (1.0x) | 11.4 ms (1.8x) | 5.2 ms (0.8x) |
| scatter_large | 134 ms | 122 ms (0.9x) | 266 ms (2.0x) | 31.5 ms (0.2x) |
| lines | 18.0 ms | 27.9 ms (1.6x) | 26.0 ms (1.4x) | 9.8 ms (0.5x) |
| polygons | 7.5 ms | 8.3 ms (1.1x) | 10.2 ms (1.4x) | 6.8 ms (0.9x) |
| text_heavy | 42.0 ms | 38.1 ms (0.9x) | 182 ms (4.3x) | 33.6 ms (0.8x) |
| ggplot_complex | 552 ms | 607 ms (1.1x) | 566 ms (1.0x) | 563 ms (1.0x) |
| Plot | base::png | ragg::agg_png | Cairo::CairoPNG | unigd png |
|---|---|---|---|---|
| scatter_small | 22.0 ms | 25.5 ms (1.2x) | 29.7 ms (1.3x) | 27.9 ms (1.3x) |
| scatter_large | 108 ms | 43.4 ms (0.4x) | 51.0 ms (0.5x) | 43.4 ms (0.4x) |
| lines | 76.4 ms | 37.6 ms (0.5x) | 53.2 ms (0.7x) | 46.3 ms (0.6x) |
| polygons | 25.4 ms | 26.1 ms (1.0x) | 26.7 ms (1.0x) | 27.5 ms (1.1x) |
| text_heavy | 167 ms | 84.5 ms (0.5x) | 49.3 ms (0.3x) | 59.5 ms (0.4x) |
| ggplot_complex | 648 ms | 617 ms (1.0x) | 628 ms (1.0x) | 611 ms (0.9x) |
| Plot | base::pdf | Cairo::CairoPDF | unigd pdf |
|---|---|---|---|
| scatter_small | 6.0 ms | 14.5 ms (2.4x) | 9.2 ms (1.5x) |
| scatter_large | 36.1 ms | 78.7 ms (2.2x) | 75.4 ms (2.1x) |
| lines | 19.1 ms | 37.8 ms (2.0x) | 30.0 ms (1.6x) |
| polygons | 6.2 ms | 11.3 ms (1.8x) | 8.4 ms (1.4x) |
| text_heavy | 3.9 ms | 35.8 ms (9.1x) | 36.9 ms (9.4x) |
| ggplot_complex | 548 ms | 584 ms (1.1x) | 601 ms (1.1x) |
| Plot | base::tiff | ragg::agg_tiff | unigd tiff |
|---|---|---|---|
| scatter_small | 13.1 ms | 6.5 ms (0.5x) | 19.5 ms (1.5x) |
| scatter_large | 95.2 ms | 20.3 ms (0.2x) | 33.9 ms (0.4x) |
| lines | 62.8 ms | 14.1 ms (0.2x) | 37.5 ms (0.6x) |
| polygons | 12.6 ms | 9.6 ms (0.8x) | 19.6 ms (1.6x) |
| text_heavy | 143 ms | 61.6 ms (0.4x) | 51.7 ms (0.4x) |
| ggplot_complex | 647 ms | 595 ms (0.9x) | 575 ms (0.9x) |
| Format | base | svglite | ragg | Cairo | unigd |
|---|---|---|---|---|---|
| SVG | 53.3 KB | 11.9 KB | — | 53.1 KB | 11.5 KB |
| PNG | 4.1 KB | — | 22.8 KB | 25.6 KB | 23.0 KB |
| 10.3 KB | — | — | 23.3 KB | 23.3 KB | |
| TIFF | 1.2 MB | — | 1.2 MB | — | 20.1 KB |
Notable: unigd TIFF output uses compression, producing files ~20x
smaller than base::tiff and ragg::agg_tiff
which default to uncompressed TIFF.
These results are specific to one machine and configuration.
base::png, Cairo::CairoPNG, unigd) perform
similarly since they share the same rasterizer.
ragg::agg_png uses its own AGG backend and can be faster on
some workloads.base::pdf uses R’s built-in PDF
writer and is consistently the fastest. unigd and
Cairo::CairoPDF both use Cairo and show similar
overhead.ragg::agg_tiff is the fastest
raster device here. unigd TIFF is slower on most plots but produces
dramatically smaller files thanks to compression.To reproduce on your own system: