Generated performance evidence

Web codec and scientific reader benchmarks.

Results stay separated by workload class. A timing is admitted only after dimensions, selection, sample or pixel hashes, output validation, and any required quality oracle pass. Unsupported and invalid rows never count as fast.

Isolated processesAbsolute peak RSSPinned fixturesGenerated claims

Web codec benchmarks

Speed, quality, and peak memory.

The 2026-08-24 profile recorded 122 validated passes and 25 explicit unsupported rows across TypeScript, explicitly registered PureJsImage WASM, other pure JavaScript libraries, jSquash WebAssembly, and Sharp/libvips.

84.7%less peak RSS than Jimp

181.8 MiBPureJsImage northstar RSS

122validated result rows

0invalid outputs or errors

The memory headline is the same 24-megapixel northstar photo workflow within one result and one environment fingerprint: 181.8 MiBfor the TypeScript path versus 1189.1 MiB for Jimp. It is not a cross-date comparison.

JPEGPNGWebPTIFFAVIF

This suite measures JPEG, PNG, WebP, TIFF, and AVIF. It does not represent every stable PureJsImage codec.

Choose a web codec benchmark chart

01Speed02Quality03Peak memory

Web codec speed

Median wall time · logarithmic scale · lower is better

Environment: Linux 6.17.0-41-generic, x64, Node v24.16.0, Intel(R) Core(TM) i7-10700 CPU @ 2.90GHz. Resize workflows use each engine’s documented default kernel.Open full-size chart

Web codec quality

Premultiplied-RGBA PSNR · higher is better

Environment: Linux 6.17.0-41-generic, x64, Node v24.16.0, Intel(R) Core(TM) i7-10700 CPU @ 2.90GHz. The quality oracle runs outside timed regions.Open full-size chart

Web codec peak memory

Absolute process peak RSS · lower is better

Environment: Linux 6.17.0-41-generic, x64, Node v24.16.0, Intel(R) Core(TM) i7-10700 CPU @ 2.90GHz. RSS includes the Node.js runtime and native allocations.Open full-size chart

Recorded web codec engines

Engine	Version	Class
purejsimage	0.16.0 (workspace)	pure-javascript
purejsimage-wasm	0.16.0 (workspace WASM)	webassembly
jimp	1.6.0	pure-javascript
sharp	0.35.3	native
sharp-single-thread	0.35.3	native-single-thread
image-js	1.7.0	pure-javascript
jsquash	avif 2.1.1; jpeg 1.6.0; png 3.1.1; webp 1.5.0; resize 2.1.1	webassembly

Current web codec benchmark report →

Scientific readers

First usable data, selected operations, source I/O, and footprint.

The 2026-08-16 PureJsImage baseline validated 43/43 workloads across 33 explicit readers. The competitor charts use only shared workloads with validated output; every published workload includes a PureJsImage row plus at least one external engine. The 2026-08-17 scaling profile admitted 6/13 PureJsImage medium/large rows under the 10% CV threshold. Unsupported semantics remain visible in chart notes and report status counts.

Environment boundary. Ordinary and scientific reports use separately fingerprinted harnesses. No cross-section speed or memory ratio is claimed.

Measured noise remains visible. Rows at or above 10% coefficient of variation are labeled directly in the timing charts and are not used as documentation headlines. See the generated JSON for the exact engine, workload, metric, and CV.

Choose a scientific reader benchmark chart

Scientific first usable block latency for shared validated scientific JS and WASM workloads — Generated from the 2026-08-16 scientific competitor artifact. Each SVG records scale direction, environment, engine class, unsupported rows, and material input-copy differences.Open full-size chart

Scientific selected-operation time for shared validated scientific JS and WASM workloads — Generated from the 2026-08-16 scientific competitor artifact. Each SVG records scale direction, environment, engine class, unsupported rows, and material input-copy differences.Open full-size chart

Absolute peak RSS

Whole isolated process, including runtime, native allocations, and WASM memory

Generated from the 2026-08-16 scientific competitor artifact. Each SVG records scale direction, environment, engine class, unsupported rows, and material input-copy differences.Open full-size chart

Scientific source I/O for shared validated scientific JS and WASM workloads — Generated from the 2026-08-16 scientific competitor artifact. Each SVG records scale direction, environment, engine class, unsupported rows, and material input-copy differences.Open full-size chart

Scientific package and WASM footprint for shared validated scientific JS and WASM workloads — Generated from the 2026-08-16 scientific competitor artifact. Each SVG records scale direction, environment, engine class, unsupported rows, and material input-copy differences.Open full-size chart

Scientific metric contract

Metric	Interpretation
Time to first usable block	Latency until validated selected data becomes available after open and initialization.
Selected-operation wall time	Complete public operation path for the pinned selection; validation remains outside timing.
Absolute peak RSS	Whole isolated process, including runtime, native allocations, and WASM memory.
Source requests and bytes	Instrumented request count, returned bytes, unique bytes, and overfetch for the selected operation.
Import and initialization	Module import, registry construction, and WASM initialization remain separate from steady-state work.
Bundle and WASM footprint	gzip/Brotli JavaScript, external or embedded WASM, installed bytes, and package count.
Correctness	Expected shape, native sample type, selected-sample hash, calibration assertions, and output bytes.

Scientific JS/WASM engine versions

Engine	Version	Class
purejsimage	0.10.0	pure-javascript
geotiff	3.0.5	pure-javascript
tiff	7.1.3	pure-javascript
utif2	4.1.0	pure-javascript
image-js	1.7.0	pure-javascript
nifti-reader-js	0.8.0	pure-javascript
npyjs	1.2.0	pure-javascript
jsfive	0.4.0	pure-javascript
h5wasm	0.10.3	webassembly
itk-wasm-image-io	1.6.1	webassembly

PureJsImage scientific baseline → · PureJsImage medium/large scaling report → · Scientific JS/WASM report → · Scientific reader tables → · Scientific format reference →

Current generated sizes

Entry points, installed packages, and WASM assets.

Minified JavaScript uses one deterministic esbuild configuration. npm package values are extracted installed bytes, not compressed tarball downloads. Native and WASM payloads are labeled separately from JavaScript wrappers.

PureJsImage major entries

Entry	Import	Minified	gzip	Brotli	Installed
Core API initial chunk	`purejsimage`	19.5 KiB	6.6 KiB	5.9 KiB	7.9 MiB
Core + scientific platform	`purejsimage/scientific`	197.5 KiB	56.3 KiB	47.4 KiB	7.9 MiB
Scientific readers: all	`purejsimage/scientific/readers/all`	1242.8 KiB	359.4 KiB	286.7 KiB	7.9 MiB
Core + common web codecs	`purejsimage/codecs/web`	651.5 KiB	238.7 KiB	198.2 KiB	7.9 MiB
Core + all stable codecs	`purejsimage/codecs/all`	1198.2 KiB	414.4 KiB	334.0 KiB	7.9 MiB

Package footprint context

The PureJsImage common-web entry is measured above. The competitor table preserves the historical JPEG/PNG-matched package targets so adding AVIF does not silently redefine the older footprint comparison; each common-web benchmark report records the exact installed engine package set it used.

Engine import	Implementation	Minified JS	Installed	Production packages
PureJsImage (matched)	pure-javascript	204.3 KiB	7.9 MiB	1
PureJsImage (all stable codecs)	pure-javascript	1198.2 KiB	7.9 MiB	1
Jimp	pure-javascript	577.4 KiB	29.3 MiB	70
image-js	pure-javascript	361.5 KiB	17.0 MiB	46
jSquash	webassembly	52.4 KiB	1.0 MiB	3
Sharp JS wrapper	native-wrapper	128.4 KiB	18.9 MiB	6

Optional PureJsImage WASM assets

Asset	Raw	gzip	Brotli
JPEG decoder WASM	31.0 KiB	7.7 KiB	6.5 KiB
JPEG decoder SIMD WASM	30.8 KiB	7.8 KiB	6.6 KiB
JPEG encoder WASM	37.1 KiB	6.5 KiB	5.3 KiB
JPEG encoder SIMD WASM	45.2 KiB	8.8 KiB	7.1 KiB
PNG codec WASM	3.8 KiB	1.6 KiB	1.4 KiB
PNG codec SIMD WASM	7.1 KiB	2.9 KiB	2.5 KiB
WebP codec WASM	9.1 KiB	3.7 KiB	3.2 KiB
WebP codec SIMD WASM	11.6 KiB	4.6 KiB	3.9 KiB

Historical deployment evidence

AWS Lambda measurements remain separately dated.

The August 9, 2026 Lambda reports remain useful for x86_64/ARM64, cold/warm, memory-tier, and explicit JPEG WASM context. They are historical deployment snapshots, not merged into the current local headline or scientific-reader charts.

ARM and explicit WASM report →· Memory-tier report →

Methodology

Reproduce and interpret the evidence.

Input bytes and fixture hashes are pinned before measurement.
Measured samples run in isolated processes after configured untimed warmups.
Output validation and independent quality measurement remain outside timed regions.
Web codec resize charts use engine-default kernels and are default-experience comparisons.
Scientific charts compare only shared operations; they do not create an all-format winner score.
Different environment fingerprints are not compared without an explicit disclaimer.

npm run fixtures:prepare
npm run fixtures:verify
npm run bench:web-codecs
npm run bench:web-codecs:charts
npm run bench:scientific:baseline
npm run bench:scientific:range
npm run bench:scientific:scaling
npm run bench:scientific:competitors:js
npm run bench:scientific:competitors:baseline
npm run bench:viewers:smoke
npm run documentation:write
npm run documentation:check

Complete benchmark methodology →· Generated result index →

Web codec and scientific reader benchmarks.

Speed, quality, and peak memory.

Web codec speed

Web codec quality

Web codec peak memory

Recorded web codec engines

First usable data, selected operations, source I/O, and footprint.

First usable block

Selected operation

Absolute peak RSS

Source I/O

Package and WASM footprint

Scientific metric contract

Scientific JS/WASM engine versions

Entry points, installed packages, and WASM assets.

PureJsImage major entries

Package footprint context

Optional PureJsImage WASM assets

AWS Lambda measurements remain separately dated.

Reproduce and interpret the evidence.