Xisto vs Node

13 Nov 2026, the xisto variants of main e5fb066 (before mandelbrot-8), on the Apple M4 and the two AWS VMs.

Setup

• Node programs: the official Node programs of the Benchmarks Game, one per problem, kept in bench/benchmarksgame/node/ (BSD-3): spectral-norm #1, n-body #8, fannkuch-redux #8, binary-trees #7, mandelbrot #1, fasta #1, k-nucleotide #8, reverse-complement #2, regex-redux #4, pidigits #2 and #3. They run on one thread.
• Runtime: Node v26.12.0 on all three machines (the official Linux tarballs on the VMs, checked against SHASUMS256.txt), run as node PROGRAM.js ARG with no flags.
• xisto: the same variants as the Rust round, built with XISTO_CPU=native. The xisto variants and what each does are listed on Xisto vs Rust.
• Command: python3 bench/benchmarksgame/langs.py 3 node=js:NODE bun=js:BUN runs both runtimes in one round, scored with python3 bench/placar.py RESULTS node (see How every run is measured).
• Steal during the round: +596 on the Xeon, +2,566 on the Graviton2 (mostly in the long Node and Bun runs).

Most xisto variants run on every core (10 on the M4, 2 on the VMs), while the Node programs run on one. The time ratio therefore mixes a faster program with more cores; the CPU columns compare the work done. The ratio in the result column is Node's time divided by xisto's.

Apple M4 (10 cores)

Source: bench/results/node-bun-mac-m4-2026-11-13-e5fb066.txt.

problem	xisto	Node	xisto time	Node time	xisto CPU	Node CPU	xisto mem	Node mem	result
spectral-norm	spectralnorm-6	spectralnorm-1	0.069 s	1.102 s	0.53 s	1.11 s	2.7 MB	46 MB	win, 16.0×
n-body	nbody	nbody-8	1.124 s	2.341 s	1.12 s	2.34 s	1.2 MB	45 MB	win, 2.1×
fannkuch-redux	fannkuchredux-5	fannkuchredux-8	1.627 s	24.913 s	15.53 s	24.91 s	3.6 MB	43 MB	win, 15.3×
binary-trees	binarytrees-2 †	binarytrees-7	0.246 s	5.194 s	1.87 s	7.00 s	120 MB	1,058 MB	win, 21.1×
mandelbrot	mandelbrot-6	mandelbrot-1	0.258 s	24.489 s	2.42 s	24.31 s	9.4 MB	50 MB	win, 94.9×
fasta	fasta	fasta-1	0.232 s	9.926 s	0.16 s	9.93 s	2.3 MB	50 MB	win, 42.8×
k-nucleotide	knucleotide-7	knucleotide-8	0.224 s	45.366 s	1.62 s	45.68 s	127 MB	704 MB	win, 202.5×
reverse-complement	revcomp-9	revcomp-2	0.124 s	1.989 s	0.10 s	2.04 s	12 MB	700 MB	win, 16.0×
regex-redux	regexredux-2	regexredux-4	0.259 s	1.168 s	0.37 s	1.23 s	242 MB	963 MB	win, 4.5×
pidigits	pidigits	pidigits-2	0.252 s	5.571 s	0.25 s	5.58 s	1.9 MB	112 MB	win, 22.1×

Xeon E5-2686 v4 (2 vCPUs)

Source: bench/results/vm-2026-11-13/node-bun-ubuntu-xeon-2026-11-13-e5fb066.txt.

problem	xisto	Node	xisto time	Node time	xisto CPU	Node CPU	xisto mem	Node mem	result
spectral-norm	spectralnorm-6	spectralnorm-1	0.459 s	3.107 s	0.89 s	3.13 s	2.3 MB	57 MB	win, 6.8×
n-body	nbody	nbody-8	2.456 s	6.228 s	2.45 s	6.23 s	2.0 MB	56 MB	win, 2.5×
fannkuch-redux	fannkuchredux-5	fannkuchredux-8	6.596 s	54.671 s	13.11 s	54.67 s	2.3 MB	54 MB	win, 8.3×
binary-trees	binarytrees †	binarytrees-7	1.624 s	34.939 s	3.05 s	51.56 s	70 MB	687 MB	win, 21.5×
mandelbrot	mandelbrot-6	mandelbrot-1	1.327 s	83.686 s	2.53 s	83.51 s	7.3 MB	59 MB	win, 63.1×
fasta									no row: Node's output differed
k-nucleotide	knucleotide-7 †	knucleotide-8	1.384 s	135.189 s	2.68 s	135.81 s	69 MB	674 MB	win, 97.7×
reverse-complement	revcomp-9	revcomp-2	0.842 s	5.301 s	0.26 s	4.73 s	11 MB	677 MB	win, 6.3×
regex-redux	regexredux-2	regexredux-4	0.690 s	3.967 s	0.97 s	4.12 s	149 MB	635 MB	win, 5.7×
pidigits	pidigits	pidigits-3	0.625 s	10.891 s	0.62 s	10.93 s	2.6 MB	121 MB	win, 17.4×

Graviton2 (2 vCPUs)

Source: bench/results/vm-2026-11-13/node-bun-arm-graviton2-2026-11-13-e5fb066.txt.

problem	xisto	Node	xisto time	Node time	xisto CPU	Node CPU	xisto mem	Node mem	result
spectral-norm	spectralnorm-7	spectralnorm-1	1.217 s	4.037 s	2.43 s	4.06 s	1.8 MB	54 MB	win, 3.3×
n-body	nbody	nbody-8	4.610 s	10.207 s	4.60 s	10.21 s	1.6 MB	54 MB	win, 2.2×
fannkuch-redux	fannkuchredux-5	fannkuchredux-8	8.575 s	67.627 s	17.13 s	67.57 s	1.8 MB	52 MB	win, 7.9×
binary-trees	binarytrees-3	binarytrees-7	1.832 s	27.839 s	3.60 s	40.02 s	114 MB	596 MB	win, 15.2×
mandelbrot	mandelbrot-7	mandelbrot-1	2.744 s	71.269 s	5.45 s	71.21 s	6.8 MB	49 MB	win, 26.0×
fasta									no row: Node's output differed
k-nucleotide	knucleotide-7	knucleotide-8	3.011 s	220.184 s	5.90 s	220.99 s	64 MB	617 MB	win, 73.1×
reverse-complement	revcomp-6 †	revcomp-2	0.282 s	4.305 s	0.24 s	4.36 s	188 MB	448 MB	win, 15.3×
regex-redux	regexredux-2	regexredux-4	0.799 s	4.746 s	1.25 s	4.95 s	149 MB	556 MB	win, 5.9×
pidigits	pidigits	pidigits-2	2.039 s	12.987 s	2.04 s	13.04 s	2.1 MB	85 MB	win, 6.4×

Notes

• Not scored: fasta on both VMs. Written to a file, the official Node fasta program's output is identical to xisto's. Written to a pipe, as the runner reads every program, it stops after about 1 MB (some 18,000 of 4.2 million lines) and exits with code 0 and no error; how much comes out varies from run to run, Node 24 does the same, and it does not happen at a tenth of the size or under strace. On the M4 the pipe output matched. The program writes one console.log per line; the cause inside Node is still being looked into.
• CPU: xisto uses less CPU than Node on every row; the largest gaps are k-nucleotide (Node's Map of strings against xisto's packed counts) and mandelbrot (scalar doubles against SIMD vectors).
• Memory: Node starts at ~45–60 MB before any work; binary-trees, k-nucleotide, reverse-complement and regex-redux reach 600 MB to 1 GB on Node.
• Before 13 Nov this page compared with Node v24 on the VMs (9 Nov) and with Node 26 on the M4 with single-threaded xisto programs (7 Nov); those files are still in bench/results/.