comparison.md

Comparison with Rapier and Avian

A comparison of oxijolt with Rapier (rapier3d) and Avian (avian3d) for people choosing a Rust physics engine: speed on Rapier's own stress scenes, the quality each engine kept while running them, what each engine offers, and where oxijolt loses. Every number in the results comes from runs whose raw files are committed under benches/comparison/results/2026-10-07-linux-5950x/; a test (benches/comparison/tests/doc_tables.rs) checks that each results row here is in that directory's summary.md and that the summary.md is what comparison summarize makes of the raw files.

Summary

Ten of Rapier's stress scenes on a 16-core Linux machine, matched settings, warm ticks, at 1, 4, 8 and 16 threads:

• Scenes of many contacts: Rapier is faster. Rapier's step is 1.9 to 2.2 times faster than oxijolt's on boxes at every thread count, 1.3 to 1.85 times on pyramid, 1.2 to 1.75 times on capsules, and 1.4 times on balls up to 8 threads (level at 16). On keva Rapier is 1.6 times faster at one thread and 1.2 times at four; oxijolt is 1.1 to 1.2 times faster at 8 and 16. On many_pyramids they are within 8 % up to four threads; oxijolt is 1.2 times faster at 8, Rapier 1.9 times at 16.
• Joints from four threads: oxijolt is faster, 1.1 to 2.1 times on joint_fixed (at every thread count), 1.1 to 1.7 times on joint_prismatic and 1.3 to 1.7 times on joint_revolute at 8 and 16 threads. At one thread Rapier is 1.2 times faster on the last two. joint_ball is the exception: Rapier is 1.1 to 1.2 times faster at one and four threads and 3.4 to 4.3 times at 8 and 16, where oxijolt's step is slower than its own four-thread step; we have not found why.
• Avian (its whole Bevy update) is the slowest of the three on every scene and thread count but two: on keva it is faster than oxijolt at one thread and level with Rapier at 8 and 16, and on joint_fixed at one thread it is level with oxijolt and ahead of Rapier.
• Quality: oxijolt and Rapier each kept 7 of the 10 scenes within the bounds (Avian 4); with each engine's defaults Rapier kept 8, oxijolt 7 and Avian 5. oxijolt lets 12 of 3008 capsules through the ground, its keva towers collapse (Avian's too), and on joint_revolute, which no engine keeps within the bounds, its joint error is 4.5 times Rapier's. On pyramid oxijolt is the only engine within the bounds: Rapier's and Avian's deepest boxes end 47 and 33 cm in the ground. Avian's ball, prismatic and revolute joints come apart.
• Fewer solver steps: with 4 velocity steps instead of Jolt's 10 (jolt-4), oxijolt is within 8 % of Rapier on pyramid from four threads and closer on boxes (Rapier 1.2 to 1.65 times faster), but leaves the bounds on five scenes where 10 steps stay within them.
• Memory: oxijolt's peak resident set is the smallest, 36 to 54 % of Rapier's at four threads.
• Determinism: every engine gave the same results at 1, 4, 8 and 16 threads, tick by tick, on every scene in both profiles.
• Defaults (each engine as shipped, sleeping on): Rapier is 5 times faster than oxijolt on balls and 8 to 9 times on boxes, most likely because its bodies fall asleep sooner (the runs count awake bodies at tick 600 only: on balls none of Rapier's and 3743 of oxijolt's).

Versions

oxijolt	1.0.0 with per-world solver step counts (WorldSettings::velocity_steps), Jolt 5.6.0, joltc 886e088
Rapier	rapier3d 0.36.0 (parry3d 0.31.1)
Avian	avian3d 0.7.0 (parry3d 0.27.0), Bevy 0.19.1
Rust	rustc 1.99.0 (b940084d7 2026-09-28), release profile, RUSTFLAGS=-C target-cpu=x86-64-v3
C++	GCC 13.3.0; Jolt and joltc built from source by CMake in Release, default features

Exact versions of every dependency are in Cargo.lock at the commit named in the results' build-info.txt.

Machine and build

An AMD Ryzen 9 5950X: 16 cores and 32 hardware threads in two complexes of 8 cores, each with its own 32 MiB L3 (CPUs 0-7 and 8-15; their SMT siblings are 16-31), boost on, AVX2 and FMA; 31 GB of memory. A rented machine: Ubuntu 24.04 in a Docker container, kernel 6.8.0-138, the schedutil frequency governor, which the container cannot change. The binaries were built on it with build_variants.py; the compiler flags of Jolt and joltc (GCC 13.3, -O3 -mavx2 -mfma and the rest) are in the results' build-info.txt, the topology in lscpu.txt and lscpu-e.txt, every dependency with its features in features.txt. The results' README says which binaries made which runs.

Earlier partial runs on a Windows desktop, one repeat with other work on the same machine, served to build the harness and are not published.

Build time and size of each engine's smallest program (hello-<engine> in benches/comparison/src/bin/), release, from an empty target directory after cargo fetch, 16 jobs, from the results' build.txt:

Build	Wall time	Executable	Stripped
oxijolt, Jolt and joltc compiled from source	18.4 s	2.54 MB	1.94 MB
oxijolt, prebuilt native library (JOLTC_LIB_DIR)	3.8 s	2.54 MB	1.94 MB
Rapier	19.6 s	3.43 MB	2.80 MB
Avian, with the Bevy it needs	42.4 s	37.06 MB	20.90 MB

MB in this table is 10^6 bytes; the peak memory columns of the results are in MiB (2^20 bytes).

Method

Scenes

The ten scenes of Rapier's examples3d/stress_tests that every engine can build, at v0.36.0, ported once to engine-neutral data (benches/comparison/src/scenes/) and built from that data by each engine's adapter. A test steps Rapier's own scene code and the Rapier adapter's world side by side and requires the same digests tick by tick (tests/rapier_oracle.rs); per-engine tests check positions, creation order, fixed or dynamic, mass (density * volume, within 1e-4), the principal moments of inertia where the engine exposes them, and shape sizes by ray casts.

Scene	Bodies	Fixed	Joints	What
balls	8000	400	0	20 x 20 x 20 balls, the bottom layer fixed, no ground
boxes	1001	1	0	ten layers of 10 x 10 cubes, each layer shifted
capsules	3009	1	0	47 layers of 8 x 8 capsules dropped 3 m into a heap
pyramid	42926	1	0	50 brick-laid layers of (50 - i)² boxes of water density, 0.55 m apart
many_pyramids	8401	1	0	40 pyramids of 20 rows
keva	38271	1	0	five stacked Keva plank towers
joint_ball	10000	26	19800	a 100 x 100 net of balls on ball joints
joint_fixed	12500	1000	20000	500 small nets on fixed joints
joint_prismatic	19200	3200	16000	3200 chains of five cubes on sliders with limits
joint_revolute	8200	200	8000	200 chains of hinged squares

Rapier's other stress scenes (heightfields, trimeshes, convex decompositions, CCD and others) are not ported.

Settings

Every run: time step 1/60 s, gravity (0, -9.81, 0), the scene's geometry, densities and creation order, sharp boxes in every engine (Jolt's boxes without convex radius), 600 ticks.

Matched profile (the headline): friction 0.5, restitution 0, no damping, no sleeping on any body, no continuous collision (Rapier max_ccd_substeps = 0, Jolt's discrete motion, no Avian SweptCcd), a 2 cm speculative contact distance in all three (Avian's NarrowPhaseConfig::default_speculative_margin), bodies joined by a joint still collide (each engine's default), no events. Jolt's mass is set to density times volume; Jolt scales its shape's inertia to it.

Solver iterations, each a row of its own:

Variant	Engine	Solver	Build
jolt	oxijolt	10 velocity and 2 position steps (Jolt's default)
jolt-4	oxijolt	4 velocity and 2 position steps
rapier-par	Rapier	4 iterations (default), 4-lane SIMD	parallel
rapier-serial	Rapier	the same	without parallel
rapier-simd8	Rapier	the same, 8-lane SIMD (tuned)	parallel, simd8
avian-par	Avian	4 substeps (as Avian's own benchmarks), joints with XPBD	parallel
avian-serial	Avian	the same	without parallel

The shared knob is iterations per step: Rapier's solver iterations, Avian's substeps, Jolt's velocity steps. Jolt keeps its position pass at the default because the other two have no such pass. A TGS substep also integrates positions, so equal counts are equal pass counts, not equal work.

Defaults profile: each engine as shipped except the time step, gravity and the scene: sleeping on with each engine's thresholds, Rapier's CCD settings, Jolt's friction 0.2 and damping 0.05, Avian's 6 substeps and unbounded speculative margin. Bodies awake at the end are reported.

Not matched, and stated: the solvers themselves (Jolt: sequential impulses with a position pass; Rapier and Avian: soft-step TGS; Avian: XPBD joints); Jolt's two broad-phase layers and its optimize_broad_phase() after building (the others build their trees in the first tick); velocity clamps (Jolt 500 m/s and 0.25 pi * 60 rad/s per body, Rapier 400 m/s); Jolt's deterministicSimulation (on); Jolt's preallocated buffers (below); Jolt shares one shape object between bodies of the same shape, Rapier and Avian create one per collider, as their scene code does; Rapier's contact recycling and clustering; friction combine rules (equal here, every body has 0.5).

Jolt's buffers: max_bodies is the body count; max_body_pairs and max_contact_constraints the smallest power of two from 2^14 at which 600 matched ticks at one thread drop no contact, doubled once; the temp allocator 1 KiB per contact constraint plus 64 MiB, doubled while 120 ticks ran over 3 % slower than with twice the size (comparison size-jolt; the table is in benches/comparison/src/engines/jolt.rs). A step that drops a contact fails the run.

Threads and pinning

Thread counts 1, 4, 8 and 16, counting the thread that steps. One thread is sequential: Jolt with a job system of concurrency 1 that leaves every job to the stepping thread; Rapier and Avian from builds without parallel. The parallel builds at one thread are rows of their own. With N threads: Jolt's thread pool with N - 1 workers plus the stepping thread; Rapier's own pool of N (configure_thread_pool, which the step installs itself) and a global Rayon pool of N; Avian's Bevy compute pool and a global Rayon pool of N, set before the app exists. A run with N threads is pinned with taskset to N logical CPUs on N different physical cores (CPUs 0 to N - 1, which lscpu -e lists as cores 0 to N - 1; their SMT siblings are 16 to 31 and stay unused). Every row reports busy cores, the process's CPU time over the wall time of the timed loop; spinning workers count as busy.

What is timed

• Jolt: PhysicsWorld::step. Rapier: PhysicsWorld::step. Avian: App::update, which also runs Bevy's schedule, time and transform propagation and Avian's systems outside its physics schedule; every update must advance Avian's physics clock by exactly one step, or the run fails. The split runs (below) time Avian's physics schedule inside the update.
• Build: Jolt's world creation, body and constraint creation and optimize_broad_phase; Rapier's insertion; Avian's app with its plugins finished and one update with the physics clock paused (spawning and collider setup). Tick 1 is the first simulated tick in all three.
• Windows of ticks: tick 1, build plus tick 1, cold (ticks 1-60: the drop and first impacts) and warm (ticks 61-600). Means, nearest-rank percentiles and maxima are over the pooled samples of the three repeats of a row; the range of the per-run warm means shows the spread between runs.
• Each run is a process of its own, from a binary built with one engine only, so feature unification between engines cannot change another engine's code and peak memory belongs to one engine. The order of the variants rotates between repeats.
• Memory: the process's peak resident set (VmHWM), read right after the last tick. It covers the process image, thread pools, Bevy, Jolt's reservations and native allocations.
• Load: before a timed run the campaign waits until other processes use under 10 % of the machine in two 5 s windows (from /proc/stat, minus the comparison's own processes), samples that load about once a minute during the run and once after it, and runs the case again (up to five attempts) when a sample passed 10 %. Every attempt is in the results' load.tsv.
• The CPU frequency governor is schedutil and boost cannot be turned off on the rented machine, which is why every timed row has three repeats and shows the spread of their means.

Quality and determinism

Separate validation runs read every body each tick (timing runs read nothing) and check bounds that separate a broken simulation from a soft one: a stack keeps at least 95 % of the height it had at tick 300 (after the drops and the pyramid's settling) and its 99th percentile speed at tick 600 is at most 0.1 m/s; no body ends more than 5 cm into the ground or more than 1 m below it; balls overlap by at most 5 cm at tick 600; joints separate by at most 10 cm, twist by at most 0.1 rad and pass a slider limit by at most 10 cm, as the per-tick 99th percentile over the joints averaged over ticks 61-600. The capsule scene is a heap, not a stack, so it has no height or final-speed bound. Peaks during impacts (deepest penetration and overlap, largest joint error) are reported next to the bounded values. A row outside a bound is shown as "below bound" next to its time, never dropped.

Each validation run also records a digest of every body's position, rotation, velocities and sleep flag per tick. The digests of 1, 4, 8 and 16 threads are compared tick by tick per engine and scene; engines are not compared with each other. A scene where no body moved faster than 0.1 m/s proves nothing and would be reported as "not moving".

Results

Linux on the 5950X above, release builds, 600 ticks per run, three repeats. The tables are copied from summary.md, which also has every quality and determinism row. Columns: warm mean, nearest-rank percentiles and maximum over the pooled ticks 61-600 of all repeats; run means is the range of the per-run warm means; cold is ticks 1-60; tick 1 and build plus tick 1 are medians over the repeats; busy cores is CPU time over wall time; peak memory is the peak resident set; quality is the validation run's verdict for the same variant, scene, profile and thread count. Times are milliseconds per tick.

Matched profile

balls

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	9.139	9.414	9.490	9.736	10.145	9.130–9.153	4.259	9.610	72.864	1.00	40	within bounds
avian-par	4	3	4.278	4.293	4.524	5.801	6.657	4.270–4.290	2.531	7.803	71.344	2.53	43	within bounds
avian-par	8	3	4.058	4.100	4.279	4.421	4.624	4.035–4.079	2.421	7.859	72.208	3.10	46	within bounds
avian-par	16	3	6.705	6.466	8.262	8.555	9.094	6.481–6.998	3.253	8.725	75.010	3.62	52	within bounds
avian-serial	1	3	8.099	8.330	8.424	8.620	8.970	8.095–8.101	3.422	8.811	70.440	1.00	39	not validated
jolt	1	3	5.135	5.338	5.699	6.094	6.216	5.107–5.166	1.660	1.064	6.062	1.00	15	within bounds
jolt	4	3	1.933	1.996	2.083	2.160	2.221	1.923–1.945	0.912	0.817	6.095	3.85	16	within bounds
jolt	8	3	1.370	1.401	1.449	1.471	1.530	1.365–1.375	0.956	0.606	5.929	7.44	17	within bounds
jolt	16	3	1.846	1.857	1.888	1.917	2.060	1.841–1.850	1.696	0.970	6.726	14.09	18	within bounds
jolt-4	1	3	4.541	4.581	5.222	5.491	5.594	4.513–4.578	1.626	1.038	6.026	1.00	15	within bounds
jolt-4	4	3	1.776	1.800	1.938	2.024	2.214	1.771–1.782	0.904	0.786	6.061	3.84	16	within bounds
jolt-4	8	3	1.280	1.298	1.353	1.382	1.427	1.277–1.283	0.962	0.745	5.978	7.39	17	within bounds
jolt-4	16	3	1.822	1.830	1.858	1.874	2.026	1.819–1.825	1.666	0.858	6.487	14.07	18	within bounds
rapier-par	1	3	3.653	3.679	4.006	4.385	5.542	3.644–3.662	2.200	8.321	12.691	1.00	37	within bounds
rapier-par	4	3	1.360	1.336	1.546	2.195	2.569	1.351–1.373	1.500	8.050	12.855	3.44	38	within bounds
rapier-par	8	3	0.968	0.945	1.143	1.618	2.065	0.950–0.978	1.425	8.191	13.159	6.10	38	within bounds
rapier-par	16	3	1.905	1.843	2.313	2.855	3.485	1.899–1.912	2.169	9.637	15.712	11.98	39	within bounds
rapier-serial	1	3	3.566	3.606	3.879	4.205	4.873	3.564–3.568	2.096	7.538	11.611	1.00	35	not validated
rapier-simd8	1	3	3.287	3.279	3.651	4.048	7.733	3.260–3.312	2.150	8.355	12.648	1.00	38	not validated
rapier-simd8	4	3	1.240	1.195	1.422	2.020	5.176	1.215–1.272	1.471	8.013	12.675	3.45	38	not validated

boxes

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	11.396	11.308	12.700	12.786	12.894	11.380–11.423	13.245	10.523	18.476	1.00	30	below bound: final_speed_p99
avian-par	4	3	6.409	6.421	7.001	7.493	7.977	6.387–6.442	7.009	7.348	15.400	2.51	32	below bound: final_speed_p99
avian-par	8	3	5.524	5.494	6.031	6.128	6.462	5.466–5.575	6.052	5.864	14.539	3.34	33	below bound: final_speed_p99
avian-par	16	3	10.016	9.933	12.362	12.659	12.932	9.414–10.388	9.389	6.000	15.387	3.67	37	below bound: final_speed_p99
avian-serial	1	3	10.852	10.804	11.986	12.062	12.164	10.837–10.863	12.602	9.947	17.102	1.00	29	not validated
jolt	1	3	6.442	6.392	6.659	7.452	8.558	6.405–6.516	8.058	42.661	43.787	1.00	13	within bounds
jolt	4	3	2.537	2.535	2.583	2.608	2.635	2.517–2.558	2.970	12.234	13.494	3.92	13	within bounds
jolt	8	3	1.806	1.805	1.832	1.846	1.865	1.796–1.816	2.016	7.192	8.550	7.70	14	within bounds
jolt	16	3	3.092	3.088	3.149	3.217	3.355	3.087–3.099	3.411	8.068	9.609	14.88	15	within bounds
jolt-4	1	3	4.127	4.109	4.278	4.558	4.629	4.108–4.157	6.926	39.801	40.931	1.00	13	within bounds
jolt-4	4	3	1.801	1.798	1.841	1.889	1.981	1.793–1.816	2.519	11.353	12.620	3.85	14	within bounds
jolt-4	8	3	1.349	1.348	1.367	1.380	1.419	1.346–1.352	1.694	6.531	7.950	7.50	14	within bounds
jolt-4	16	3	2.348	2.363	2.495	2.512	2.526	2.223–2.453	2.602	6.959	8.505	14.59	15	within bounds
rapier-par	1	3	3.326	3.310	3.413	3.459	3.592	3.303–3.362	3.693	23.352	24.243	1.00	25	within bounds
rapier-par	4	3	1.152	1.147	1.186	1.203	1.505	1.139–1.169	1.346	12.392	13.681	3.90	25	within bounds
rapier-par	8	3	0.816	0.802	0.961	1.022	1.199	0.757–0.891	1.141	12.393	13.873	7.58	26	within bounds
rapier-par	16	3	1.606	1.544	1.854	1.900	2.439	1.502–1.797	1.900	15.692	18.039	14.99	26	within bounds
rapier-serial	1	3	3.271	3.277	3.328	3.402	3.542	3.230–3.299	3.583	19.319	19.997	1.00	24	not validated
rapier-simd8	1	3	2.616	2.560	2.755	2.874	3.470	2.550–2.739	3.031	23.081	23.913	0.99	25	not validated
rapier-simd8	4	3	0.926	0.924	0.947	0.953	0.965	0.914–0.938	1.127	12.295	13.401	3.92	25	not validated

capsules

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	10.119	11.281	13.436	13.538	13.678	10.072–10.176	3.079	7.167	28.648	1.00	35	within bounds
avian-par	4	3	5.507	5.787	7.334	9.079	10.140	5.496–5.517	1.660	5.644	27.217	2.39	37	within bounds
avian-par	8	3	4.571	5.024	5.947	6.091	6.370	4.552–4.594	1.480	5.350	27.503	3.24	39	within bounds
avian-par	16	3	7.420	7.731	11.374	11.677	11.944	7.018–7.679	1.969	5.877	29.740	3.83	44	within bounds
avian-serial	1	3	9.303	10.410	12.669	12.894	16.762	9.286–9.318	2.497	6.393	27.231	1.00	34	not validated
jolt	1	3	8.107	10.251	11.150	11.417	11.586	8.042–8.161	1.014	1.067	3.347	1.00	13	below bound: ground_penetration, fallen
jolt	4	3	2.915	3.624	3.965	4.053	4.080	2.910–2.919	0.479	0.585	3.142	3.74	14	below bound: ground_penetration, fallen
jolt	8	3	2.006	2.460	2.706	2.752	2.845	2.002–2.012	0.366	0.426	2.899	7.07	14	below bound: ground_penetration, fallen
jolt	16	3	2.928	3.416	4.001	4.104	4.210	2.878–2.999	0.668	0.587	3.517	13.66	16	below bound: ground_penetration, fallen
jolt-4	1	3	8.306	10.495	12.139	12.394	12.540	8.274–8.358	0.967	1.034	3.311	1.00	15	below bound: ground_penetration, fallen
jolt-4	4	3	3.037	3.847	4.328	4.425	4.472	3.029–3.043	0.468	0.602	3.108	3.74	15	below bound: ground_penetration, fallen
jolt-4	8	3	2.108	2.644	2.995	3.034	3.083	2.104–2.114	0.360	0.441	3.046	7.10	16	below bound: ground_penetration, fallen
jolt-4	16	3	2.901	3.514	3.963	4.130	4.242	2.891–2.909	0.639	0.564	3.473	13.56	17	below bound: ground_penetration, fallen
rapier-par	1	3	4.630	5.141	6.063	6.167	6.460	4.619–4.645	1.127	6.290	8.481	1.00	37	within bounds
rapier-par	4	3	1.775	1.929	2.311	2.386	2.792	1.770–1.784	0.801	5.033	7.704	3.59	39	within bounds
rapier-par	8	3	1.312	1.392	1.704	1.756	2.066	1.305–1.319	0.772	4.911	7.827	6.52	40	within bounds
rapier-par	16	3	2.431	2.588	3.082	3.182	3.392	2.391–2.478	1.182	5.870	9.664	11.72	40	within bounds
rapier-serial	1	3	4.524	5.016	5.935	6.140	9.541	4.478–4.586	1.041	5.314	7.250	1.00	36	not validated
rapier-simd8	1	3	4.392	4.872	5.820	5.965	7.286	4.379–4.410	1.110	6.274	8.435	1.00	38	not validated
rapier-simd8	4	3	1.713	1.837	2.246	2.332	3.388	1.703–1.731	0.814	4.970	7.602	3.58	39	not validated

pyramid

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	589.059	600.149	615.133	621.292	683.607	588.179–589.521	103.394	47.844	889.391	1.00	285	below bound: ground_penetration, final_speed_p99
avian-par	4	3	231.822	235.022	252.520	257.242	308.585	230.385–232.831	43.800	36.086	888.149	2.78	329	below bound: ground_penetration, final_speed_p99
avian-par	8	3	165.067	168.955	179.960	183.382	209.460	164.714–165.466	33.418	35.029	893.298	4.39	314	below bound: ground_penetration, final_speed_p99
avian-par	16	3	161.380	164.042	176.319	180.898	239.327	160.664–162.119	38.698	37.276	892.577	6.03	349	below bound: ground_penetration, final_speed_p99
avian-serial	1	3	582.178	592.939	607.578	621.007	679.104	581.492–582.690	99.945	43.803	875.095	1.00	283	not validated
jolt	1	3	473.674	473.264	535.147	542.644	632.148	473.259–474.348	73.989	8.825	39.153	1.00	183	within bounds
jolt	4	3	173.360	174.863	189.249	191.437	238.574	172.722–173.841	24.860	5.132	36.348	3.90	188	within bounds
jolt	8	3	122.948	125.456	130.338	132.714	188.668	122.814–123.056	16.975	3.555	34.915	7.66	192	within bounds
jolt	16	3	119.346	122.987	125.867	133.791	171.015	118.135–120.037	18.782	4.606	36.512	15.07	202	within bounds
jolt-4	1	3	311.008	298.841	402.265	427.769	520.825	310.945–311.128	69.585	8.796	38.892	1.00	185	below bound: final_speed_p99
jolt-4	4	3	119.712	117.962	141.763	147.872	210.531	119.424–120.053	23.011	5.222	36.367	3.85	191	below bound: final_speed_p99
jolt-4	8	3	87.224	87.602	97.595	100.940	154.940	87.012–87.450	15.850	3.615	33.669	7.51	194	below bound: final_speed_p99
jolt-4	16	3	86.186	87.869	92.118	98.741	132.972	86.008–86.349	16.984	4.271	35.631	14.71	203	below bound: final_speed_p99
rapier-par	1	3	256.213	261.979	265.904	268.976	345.855	255.866–256.562	49.858	57.007	85.878	1.00	482	below bound: ground_penetration
rapier-par	4	3	113.148	115.684	117.408	119.672	168.635	113.075–113.203	24.705	54.474	83.352	3.80	484	below bound: ground_penetration
rapier-par	8	3	94.378	96.614	97.730	98.964	99.831	94.155–94.540	23.579	55.231	84.522	7.41	483	below bound: ground_penetration
rapier-par	16	3	81.097	82.637	86.900	93.468	134.393	80.740–81.407	22.186	59.737	90.634	14.37	485	below bound: ground_penetration
rapier-serial	1	3	251.521	256.666	262.454	274.853	338.982	250.481–253.352	48.457	53.319	81.907	1.00	474	not validated
rapier-simd8	1	3	245.006	250.153	254.612	257.748	315.688	244.460–245.941	49.222	56.781	85.524	1.00	482	not validated
rapier-simd8	4	3	102.509	104.389	106.900	111.925	186.501	102.240–102.735	24.553	54.106	82.575	3.77	483	not validated

many_pyramids

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	69.058	68.825	70.214	74.065	87.059	68.665–69.551	65.710	35.330	104.644	1.00	65	within bounds
avian-par	4	3	27.086	26.928	30.439	30.742	33.326	26.801–27.424	23.709	21.504	92.458	2.90	70	within bounds
avian-par	8	3	17.669	17.632	18.324	18.855	20.741	17.429–17.848	17.229	17.718	88.717	4.74	70	within bounds
avian-par	16	3	27.035	25.382	31.490	31.726	32.044	24.789–28.637	25.992	16.890	89.406	5.50	82	within bounds
avian-serial	1	3	66.815	66.645	67.504	71.652	75.901	66.718–66.869	63.501	33.552	101.680	1.00	63	not validated
jolt	1	3	25.145	25.069	25.358	27.236	43.788	24.950–25.357	27.524	61.454	66.981	1.00	33	within bounds
jolt	4	3	8.451	8.450	8.506	8.527	9.118	8.430–8.476	9.027	19.953	25.673	3.95	34	within bounds
jolt	8	3	5.245	5.243	5.293	5.310	5.537	5.235–5.253	5.454	12.553	18.480	7.82	35	within bounds
jolt	16	3	9.419	9.417	10.252	10.770	11.237	9.193–9.628	9.590	14.366	20.474	15.56	37	within bounds
jolt-4	1	3	17.562	17.477	18.135	20.153	22.761	17.373–17.663	25.650	51.303	56.940	1.00	33	below bound: final_speed_p99
jolt-4	4	3	6.076	6.067	6.240	6.783	7.429	6.026–6.115	8.025	17.015	22.897	3.93	34	below bound: final_speed_p99
jolt-4	8	3	3.916	3.902	4.158	4.319	4.635	3.862–3.989	4.799	11.176	17.729	7.75	35	below bound: final_speed_p99
jolt-4	16	3	7.163	7.372	7.516	7.551	7.604	6.731–7.395	7.438	12.534	18.702	15.38	37	below bound: final_speed_p99
rapier-par	1	3	23.231	23.150	24.329	24.973	25.786	23.129–23.338	24.777	97.384	104.555	1.00	86	within bounds
rapier-par	4	3	8.418	8.413	8.577	8.717	8.939	8.372–8.464	9.145	53.811	61.338	3.73	86	within bounds
rapier-par	8	3	6.272	6.267	6.364	6.457	8.647	6.262–6.277	7.019	51.355	58.845	7.11	87	within bounds
rapier-par	16	3	4.917	4.915	5.262	5.941	6.457	4.732–5.077	5.823	56.076	64.621	13.33	87	within bounds
rapier-serial	1	3	23.624	23.633	24.885	25.292	25.931	23.515–23.759	24.482	82.688	89.279	1.00	83	not validated
rapier-simd8	1	3	21.880	21.756	23.604	24.539	25.060	21.654–22.112	23.089	96.725	103.666	1.00	86	not validated
rapier-simd8	4	3	7.804	7.790	7.988	8.128	8.395	7.791–7.819	8.514	53.476	60.628	3.70	86	not validated

keva

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	506.118	587.790	676.791	686.943	742.031	504.161–509.328	643.068	314.315	1029.065	1.00	434	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
avian-par	4	3	195.277	219.053	246.462	250.315	277.198	194.847–195.670	248.468	152.132	856.561	2.64	503	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
avian-par	8	3	132.520	145.520	166.926	171.477	221.550	132.153–132.918	170.470	118.354	835.955	4.18	478	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
avian-par	16	3	121.566	131.401	153.826	161.352	216.728	121.132–122.048	167.164	131.848	856.240	5.96	488	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
avian-serial	1	3	496.560	576.022	663.304	676.732	780.875	494.907–497.501	634.465	306.636	1011.842	1.00	432	not validated
jolt	1	3	577.070	592.871	834.559	850.231	954.532	576.249–578.468	814.740	975.060	1003.044	1.00	240	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
jolt	4	3	191.701	199.664	276.785	281.186	367.747	191.618–191.862	283.819	348.684	377.902	3.93	245	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
jolt	8	3	122.124	127.704	177.464	186.316	222.705	121.915–122.305	195.172	244.070	272.519	7.74	248	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
jolt	16	3	102.088	107.713	151.332	161.700	192.268	101.633–102.760	182.778	276.332	305.328	15.18	254	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
jolt-4	1	3	745.867	767.674	1177.652	1225.775	1290.947	745.069–746.837	589.750	743.277	771.035	1.00	303	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
jolt-4	4	3	241.447	247.111	377.131	395.977	401.437	240.756–241.877	206.025	271.120	299.506	3.94	307	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
jolt-4	8	3	153.205	154.975	240.089	252.914	257.238	152.958–153.472	142.536	189.660	218.392	7.76	309	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
jolt-4	16	3	125.429	125.735	196.074	207.295	219.445	125.055–125.854	132.565	219.005	248.089	15.23	316	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
rapier-par	1	3	352.861	352.424	360.251	368.786	451.404	351.938–354.373	371.938	1097.665	1125.667	1.00	678	below bound: final_speed_p99
rapier-par	4	3	164.183	163.870	166.611	173.786	249.092	163.861–164.405	173.160	584.109	612.309	3.83	681	below bound: final_speed_p99
rapier-par	8	3	137.334	137.155	138.977	141.168	208.801	137.068–137.481	145.428	537.131	566.002	7.51	687	below bound: final_speed_p99
rapier-par	16	3	122.105	121.886	124.546	129.483	197.895	121.761–122.489	130.964	572.200	601.847	14.73	692	below bound: final_speed_p99
rapier-serial	1	3	356.382	355.403	365.194	374.492	454.193	353.032–360.010	370.593	925.590	953.111	1.00	663	not validated
rapier-simd8	1	3	343.351	343.572	350.024	354.437	444.261	342.408–344.662	363.003	1060.282	1088.076	1.00	678	not validated
rapier-simd8	4	3	146.720	146.602	149.043	150.124	157.087	146.540–147.043	155.831	569.968	597.554	3.81	680	not validated

joint_ball

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	21.046	13.354	50.248	64.374	69.131	20.952–21.102	11.470	17.915	125.846	1.00	80	below bound: anchor_p99
avian-par	4	3	15.209	10.691	32.674	41.187	47.278	15.126–15.285	9.330	14.633	123.471	1.62	87	below bound: anchor_p99
avian-par	8	3	13.859	10.180	28.195	35.756	39.126	13.848–13.882	9.021	13.769	122.879	1.90	89	below bound: anchor_p99
avian-par	16	3	16.895	11.775	35.926	44.367	48.076	16.334–17.204	10.446	15.912	126.986	2.43	92	below bound: anchor_p99
avian-serial	1	3	20.060	12.374	48.676	62.580	64.663	19.844–20.285	10.516	16.508	121.298	1.00	79	not validated
jolt	1	3	11.364	11.233	11.868	13.132	25.938	11.237–11.592	10.322	7.826	225.895	1.00	26	within bounds
jolt	4	3	4.509	4.511	4.642	4.695	4.850	4.490–4.543	4.252	4.617	225.456	3.42	27	within bounds
jolt	8	3	10.677	10.217	17.188	18.393	19.074	8.325–12.128	3.776	4.116	217.150	6.23	28	within bounds
jolt	16	3	15.395	15.585	18.088	18.782	20.527	15.172–15.606	11.962	6.100	227.830	11.26	30	within bounds
jolt-4	1	3	9.416	9.366	9.951	10.063	10.717	9.336–9.505	8.212	6.534	225.375	1.00	26	below bound: anchor_p99
jolt-4	4	3	3.771	3.763	3.948	4.015	4.367	3.741–3.811	3.477	3.654	224.446	3.32	27	below bound: anchor_p99
jolt-4	8	3	12.676	14.209	16.249	16.763	17.580	12.221–13.018	2.920	3.285	216.099	5.80	28	below bound: anchor_p99
jolt-4	16	3	8.277	5.355	15.987	16.768	18.202	5.145–14.520	7.271	5.133	229.973	10.81	30	below bound: anchor_p99
rapier-par	1	3	10.508	10.364	11.518	11.769	12.971	10.441–10.562	10.388	34.672	48.544	1.00	61	within bounds
rapier-par	4	3	3.624	3.575	3.969	4.052	7.998	3.585–3.680	3.910	29.204	42.923	3.50	62	within bounds
rapier-par	8	3	2.492	2.467	2.722	2.782	2.964	2.482–2.497	3.176	27.946	42.356	6.23	63	within bounds
rapier-par	16	3	4.515	4.510	4.770	4.867	5.111	4.463–4.545	5.102	32.027	47.509	12.92	64	within bounds
rapier-serial	1	3	10.053	9.941	11.002	11.211	11.862	10.041–10.060	10.019	34.081	47.430	1.00	59	not validated
rapier-simd8	1	3	8.808	8.691	9.816	10.024	12.413	8.728–8.891	8.619	33.180	46.895	1.00	61	not validated
rapier-simd8	4	3	3.150	3.120	3.485	3.581	3.803	3.146–3.156	3.404	28.925	42.549	3.45	62	not validated

joint_fixed

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	15.449	15.399	16.069	16.299	16.732	15.440–15.457	15.673	23.057	144.579	1.00	53	within bounds
avian-par	4	3	13.843	13.842	14.247	14.589	14.952	13.819–13.883	13.842	20.212	145.355	1.23	54	within bounds
avian-par	8	3	13.588	13.612	13.934	14.040	14.408	13.574–13.598	13.665	20.317	142.458	1.29	55	within bounds
avian-par	16	3	15.218	15.151	15.951	16.105	16.272	15.038–15.418	15.135	23.126	149.500	1.47	58	within bounds
avian-serial	1	3	14.550	14.523	15.140	15.397	16.463	14.538–14.560	14.774	21.597	141.292	1.00	51	not validated
jolt	1	3	15.436	15.433	15.542	16.091	17.693	15.354–15.484	15.351	14.864	242.356	1.00	31	within bounds
jolt	4	3	4.361	4.359	4.391	4.412	5.096	4.350–4.370	4.385	4.877	234.738	3.90	32	within bounds
jolt	8	3	2.361	2.362	2.392	2.404	2.520	2.352–2.369	2.404	2.863	232.830	7.62	32	within bounds
jolt	16	3	2.743	2.691	3.024	3.093	3.328	2.735–2.752	2.776	3.205	232.830	11.64	34	within bounds
jolt-4	1	3	12.939	12.922	13.056	13.467	14.679	12.890–12.968	12.933	12.204	240.076	1.00	31	below bound: angle_p99
jolt-4	4	3	3.710	3.710	3.747	3.763	4.312	3.694–3.719	3.756	4.062	230.776	3.89	32	below bound: angle_p99
jolt-4	8	3	2.005	2.004	2.021	2.041	2.194	2.004–2.006	2.062	2.426	225.052	7.56	33	below bound: angle_p99
jolt-4	16	3	2.762	2.719	3.063	3.163	3.264	2.729–2.827	2.814	3.291	235.475	10.73	36	below bound: angle_p99
rapier-par	1	3	17.074	17.035	17.486	17.946	19.709	17.040–17.121	17.732	50.355	64.763	1.00	62	within bounds
rapier-par	4	3	5.799	5.764	6.009	6.499	7.010	5.789–5.811	6.518	39.625	54.209	3.64	63	within bounds
rapier-par	8	3	3.887	3.856	4.099	4.624	4.854	3.831–3.949	4.720	37.929	53.429	6.72	63	within bounds
rapier-par	16	3	5.785	5.763	6.227	6.609	7.205	5.661–5.922	6.664	42.704	58.805	14.08	64	within bounds
rapier-serial	1	3	16.817	16.785	17.178	17.551	19.692	16.774–16.865	17.640	49.248	63.277	1.00	61	not validated
rapier-simd8	1	3	13.350	13.265	14.003	14.651	16.433	13.207–13.554	14.428	47.219	61.514	1.00	62	not validated
rapier-simd8	4	3	4.855	4.818	5.076	5.727	7.567	4.836–4.885	5.779	38.914	53.491	3.57	63	not validated

joint_prismatic

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	50.235	54.261	55.854	56.442	59.967	50.167–50.339	18.233	29.162	241.615	1.00	80	below bound: anchor_p99, angle_p99, limit_p99
avian-par	4	3	31.798	33.486	35.384	35.949	36.661	31.707–31.924	14.568	25.149	242.423	1.82	91	below bound: anchor_p99, angle_p99, limit_p99
avian-par	8	3	27.835	29.139	30.672	31.459	33.772	27.639–28.224	14.097	24.882	241.588	2.26	91	below bound: anchor_p99, angle_p99, limit_p99
avian-par	16	3	33.957	34.716	40.341	41.648	42.254	33.704–34.457	16.526	30.084	250.517	2.76	103	below bound: anchor_p99, angle_p99, limit_p99
avian-serial	1	3	48.336	52.193	53.846	54.839	57.212	48.145–48.485	16.874	27.294	240.148	1.00	79	not validated
jolt	1	3	18.779	18.959	19.250	19.918	24.181	18.675–18.942	15.898	14.803	149.780	1.00	36	within bounds
jolt	4	3	5.709	5.690	6.187	7.874	10.264	5.611–5.884	4.902	5.170	142.325	3.91	37	within bounds
jolt	8	3	3.085	3.101	3.207	3.240	3.462	3.063–3.110	2.680	3.162	140.518	7.68	39	within bounds
jolt	16	3	3.909	3.917	4.103	4.177	4.263	3.879–3.938	3.643	4.516	143.922	11.77	42	within bounds
jolt-4	1	3	15.990	16.130	16.348	16.896	20.901	15.851–16.075	14.147	12.340	146.338	1.00	37	below bound: anchor_p99
jolt-4	4	3	4.890	4.929	5.000	5.032	5.304	4.877–4.909	4.445	4.463	137.167	3.90	38	below bound: anchor_p99
jolt-4	8	3	2.714	2.736	2.798	2.824	2.883	2.705–2.720	2.473	2.937	138.833	7.63	39	below bound: anchor_p99
jolt-4	16	3	3.837	3.841	4.082	4.178	4.392	3.817–3.852	3.673	3.762	141.769	11.18	42	below bound: anchor_p99
rapier-par	1	3	15.997	15.811	16.766	17.440	20.605	15.630–16.424	17.253	50.420	70.393	1.00	76	within bounds
rapier-par	4	3	6.112	6.103	6.516	7.074	8.789	6.077–6.150	7.143	41.072	61.563	3.49	77	within bounds
rapier-par	8	3	5.319	5.156	6.450	6.992	8.504	5.249–5.399	7.415	38.872	59.803	6.36	78	within bounds
rapier-par	16	3	6.291	6.241	7.050	7.558	8.754	6.150–6.466	8.483	42.935	65.083	12.57	79	within bounds
rapier-serial	1	3	15.686	15.853	16.338	16.904	20.686	15.383–15.842	17.115	48.920	68.465	1.00	74	not validated
rapier-simd8	1	3	13.799	13.896	14.419	15.076	19.580	13.710–13.882	14.952	47.609	67.373	1.00	76	not validated
rapier-simd8	4	3	5.426	5.415	5.823	6.476	8.131	5.387–5.457	6.442	40.297	60.837	3.43	77	not validated

joint_revolute

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	13.470	13.338	14.704	15.343	15.655	13.424–13.540	9.191	12.415	81.736	1.00	43	below bound: anchor_p99, angle_p99
avian-par	4	3	9.844	9.775	10.595	10.893	11.087	9.823–9.861	7.275	10.401	80.674	1.58	45	below bound: anchor_p99, angle_p99
avian-par	8	3	9.177	9.135	9.775	10.026	10.311	9.144–9.195	7.129	10.198	80.011	1.78	46	below bound: anchor_p99, angle_p99
avian-par	16	3	10.874	10.699	11.902	12.749	13.794	10.577–11.386	8.253	12.253	82.176	2.10	50	below bound: anchor_p99, angle_p99
avian-serial	1	3	12.483	12.276	13.916	14.355	21.908	12.388–12.617	8.293	11.539	79.990	1.00	41	not validated
jolt	1	3	7.952	7.936	8.493	8.812	15.837	7.852–8.112	6.787	4.231	28.969	1.00	22	below bound: anchor_p99, angle_p99
jolt	4	3	2.275	2.269	2.459	2.792	2.858	2.272–2.279	2.007	1.564	26.883	3.90	22	below bound: anchor_p99, angle_p99
jolt	8	3	1.254	1.251	1.348	1.605	1.659	1.252–1.256	1.106	1.091	26.236	7.55	23	below bound: anchor_p99, angle_p99
jolt	16	3	1.962	1.961	2.113	2.497	2.679	1.957–1.965	1.727	1.165	26.985	10.47	25	below bound: anchor_p99, angle_p99
jolt-4	1	3	6.539	6.546	7.199	7.426	7.500	6.528–6.546	5.291	3.706	28.264	1.00	22	below bound: anchor_p99, angle_p99
jolt-4	4	3	1.907	1.888	2.210	2.326	2.377	1.901–1.911	1.581	1.421	26.184	3.89	22	below bound: anchor_p99, angle_p99
jolt-4	8	3	1.115	1.116	1.224	1.307	1.337	1.109–1.120	0.944	1.027	26.191	7.13	23	below bound: anchor_p99, angle_p99
jolt-4	16	3	1.977	1.966	2.219	2.493	2.676	1.951–2.024	1.826	1.195	27.141	9.88	25	below bound: anchor_p99, angle_p99
rapier-par	1	3	6.634	6.432	7.718	7.837	7.944	6.619–6.655	7.513	20.101	30.050	1.00	42	below bound: anchor_p99, angle_p99
rapier-par	4	3	2.329	2.255	3.138	3.218	3.285	2.327–2.333	2.688	16.956	27.065	3.54	42	below bound: anchor_p99, angle_p99
rapier-par	8	3	1.609	1.526	2.460	2.542	2.602	1.604–1.615	2.102	16.262	26.891	6.37	43	below bound: anchor_p99, angle_p99
rapier-par	16	3	3.270	3.208	4.311	4.504	4.828	3.209–3.333	3.224	19.105	30.753	12.62	44	below bound: anchor_p99, angle_p99
rapier-serial	1	3	6.555	6.326	7.609	7.811	8.299	6.513–6.623	7.865	19.351	28.913	1.00	40	not validated
rapier-simd8	1	3	5.463	5.289	6.650	6.808	11.361	5.449–5.475	6.075	19.081	28.908	1.00	46	not validated
rapier-simd8	4	3	1.993	1.908	2.848	2.920	7.260	1.987–1.999	2.269	16.644	26.800	3.50	47	not validated

Defaults profile

balls

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	6.140	6.527	11.754	11.875	13.205	6.129–6.155	4.831	9.945	69.931	1.00	47	within bounds
avian-par	4	3	3.560	3.962	5.666	6.418	7.920	3.548–3.578	3.023	8.051	69.226	2.31	49	within bounds
jolt	1	3	4.667	5.354	5.713	6.100	6.320	4.637–4.689	1.727	1.115	6.109	1.00	15	within bounds
jolt	4	3	1.749	2.011	2.093	2.188	2.235	1.745–1.751	0.965	0.892	6.127	3.85	16	within bounds
rapier-par	1	3	0.900	0.011	4.271	4.605	5.465	0.894–0.906	2.472	8.512	12.783	0.99	37	within bounds
rapier-par	4	3	0.355	0.008	1.606	2.163	2.657	0.354–0.356	1.633	8.352	13.133	3.02	38	within bounds

boxes

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	14.904	15.079	15.853	16.815	17.958	14.862–14.951	16.229	10.684	18.103	1.00	30	within bounds
avian-par	4	3	8.291	8.319	8.941	9.989	10.857	8.272–8.305	8.905	7.468	15.271	2.54	33	within bounds
jolt	1	3	2.483	0.080	6.645	6.915	7.025	2.478–2.487	8.376	42.632	43.757	1.00	13	within bounds
jolt	4	3	0.982	0.054	2.586	2.636	2.669	0.980–0.983	3.024	12.211	13.634	3.85	13	within bounds
rapier-par	1	3	0.303	0.010	3.297	3.312	3.376	0.303–0.304	3.710	23.359	24.184	0.98	25	within bounds
rapier-par	4	3	0.107	0.005	1.136	1.159	1.170	0.105–0.108	1.350	12.429	14.074	3.75	25	within bounds

capsules

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	13.127	15.456	16.983	17.306	17.719	13.114–13.133	3.416	7.223	27.508	1.00	34	within bounds
avian-par	4	3	7.395	8.038	9.594	10.357	11.841	7.364–7.430	1.952	5.795	26.208	2.42	37	within bounds
jolt	1	3	7.699	9.268	10.913	11.107	11.163	7.683–7.717	1.015	1.056	3.373	1.00	13	below bound: ground_penetration, fallen
jolt	4	3	2.799	3.380	3.960	4.031	4.080	2.795–2.805	0.487	0.584	3.012	3.72	14	below bound: ground_penetration, fallen
rapier-par	1	3	4.878	5.380	6.340	6.499	7.157	4.834–4.930	1.261	6.430	8.573	1.00	38	within bounds
rapier-par	4	3	1.919	2.054	2.439	4.188	5.947	1.853–2.042	0.893	5.054	7.520	3.57	39	within bounds

pyramid

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	751.793	766.411	801.873	818.228	859.866	750.839–753.386	127.599	50.647	867.627	1.00	282	below bound: final_speed_p99
avian-par	4	3	291.128	290.706	328.329	333.217	394.826	290.490–292.379	53.253	38.680	861.652	2.80	330	below bound: final_speed_p99
jolt	1	3	462.700	459.215	528.140	558.399	625.164	460.443–464.080	75.043	9.419	40.258	1.00	183	within bounds
jolt	4	3	170.530	171.391	188.142	194.661	260.032	169.982–171.297	24.887	5.164	35.184	3.89	189	within bounds
rapier-par	1	3	256.655	262.348	266.850	274.423	352.225	255.559–257.335	50.431	57.920	86.917	1.00	483	within bounds
rapier-par	4	3	114.092	116.463	118.871	123.763	204.279	113.527–114.683	24.750	56.251	85.535	3.79	485	within bounds

many_pyramids

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	76.845	76.138	82.302	87.081	91.297	76.710–76.954	81.588	35.556	100.902	1.00	65	within bounds
avian-par	4	3	28.669	28.114	32.232	32.938	35.030	28.556–28.845	29.509	20.549	86.778	3.10	70	within bounds
jolt	1	3	0.001	0.001	0.001	0.001	0.004	0.001–0.001	18.566	61.933	67.532	1.00	33	within bounds
jolt	4	3	0.001	0.001	0.001	0.001	0.005	0.001–0.001	6.085	20.020	25.825	3.95	34	within bounds
rapier-par	1	3	0.011	0.011	0.011	0.013	0.016	0.011–0.011	21.911	98.084	104.907	1.00	86	within bounds
rapier-par	4	3	0.007	0.006	0.013	0.015	0.026	0.006–0.008	8.178	52.779	60.424	3.57	86	within bounds

keva

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	718.809	784.663	919.021	941.957	1019.130	718.075–719.940	842.786	322.174	1015.143	1.00	428	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
avian-par	4	3	272.581	282.627	371.613	386.609	460.744	271.662–273.474	318.026	154.663	863.631	2.68	496	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
jolt	1	3	568.743	574.171	805.897	834.828	940.049	566.907–570.180	818.127	970.693	998.586	1.00	240	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
jolt	4	3	184.082	184.548	260.093	270.477	282.408	183.772–184.451	283.830	348.749	376.627	3.92	245	below bound: height_ratio, ground_penetration, fallen, final_speed_p99
rapier-par	1	3	352.573	352.222	359.616	365.454	463.484	350.384–354.233	371.626	1093.099	1121.023	1.00	680	below bound: final_speed_p99
rapier-par	4	3	163.581	163.352	166.679	168.331	175.419	162.759–164.911	173.842	587.928	616.348	3.83	682	below bound: final_speed_p99

joint_ball

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	17.941	18.071	19.618	19.729	20.672	17.901–17.967	14.684	19.657	123.598	1.00	55	below bound: anchor_p99
avian-par	4	3	14.174	14.144	15.109	15.326	16.079	14.153–14.191	12.272	15.525	117.706	1.42	55	below bound: anchor_p99
jolt	1	3	11.442	11.465	11.805	12.097	12.736	11.433–11.460	10.412	7.994	226.044	1.00	26	within bounds
jolt	4	3	4.615	4.619	4.762	4.826	4.894	4.608–4.620	4.394	4.490	224.171	3.39	27	within bounds
rapier-par	1	3	11.457	11.449	12.613	12.822	13.229	11.420–11.491	10.835	37.624	51.220	1.00	62	within bounds
rapier-par	4	3	3.920	3.911	4.301	4.393	4.715	3.895–3.952	4.065	32.062	46.233	3.51	63	within bounds

joint_fixed

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	20.269	20.246	20.785	21.108	21.730	20.221–20.315	20.520	27.577	145.599	1.00	51	within bounds
avian-par	4	3	18.759	18.726	19.457	20.445	25.634	18.657–18.817	18.710	24.167	140.753	1.19	52	within bounds
jolt	1	3	15.522	15.515	15.616	15.989	17.427	15.470–15.560	15.469	14.977	242.026	1.00	31	within bounds
jolt	4	3	4.399	4.399	4.428	4.440	4.462	4.389–4.409	4.448	4.868	236.841	3.90	32	within bounds
rapier-par	1	3	8.913	17.019	17.972	18.459	20.593	8.846–8.950	18.281	53.171	67.447	1.00	62	within bounds
rapier-par	4	3	3.018	5.748	6.040	6.476	6.849	2.970–3.046	6.632	42.346	56.859	3.61	63	within bounds

joint_prismatic

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	58.168	58.903	73.049	74.370	77.913	58.146–58.193	23.836	33.593	239.173	1.00	77	below bound: anchor_p99, angle_p99, limit_p99
avian-par	4	3	37.092	37.461	44.039	45.113	46.438	37.074–37.106	19.467	29.715	239.951	1.83	88	below bound: anchor_p99, angle_p99, limit_p99
jolt	1	3	18.899	19.101	19.294	19.786	22.564	18.850–18.959	15.986	15.049	150.176	1.00	36	within bounds
jolt	4	3	5.668	5.725	5.794	5.815	6.183	5.653–5.680	4.930	5.147	142.504	3.92	37	within bounds
rapier-par	1	3	16.800	16.929	17.427	18.142	22.425	16.462–16.983	18.044	52.286	72.042	1.00	77	within bounds
rapier-par	4	3	6.287	6.272	6.711	7.299	8.956	6.234–6.352	7.305	43.244	63.261	3.49	77	within bounds

joint_revolute

variant	threads	runs	warm mean ms	warm p50	warm p95	warm p99	warm max	run means	cold mean	tick 1	build + tick 1	busy cores	peak memory MiB	quality
avian-par	1	3	65.498	64.936	70.929	78.138	80.724	65.414–65.568	19.538	13.566	80.010	1.00	63	below bound: anchor_p99, angle_p99
avian-par	4	3	38.553	38.204	42.187	45.392	65.563	38.349–38.701	14.112	11.906	79.154	1.98	73	below bound: anchor_p99, angle_p99
jolt	1	3	7.890	7.924	8.291	8.495	8.676	7.874–7.916	6.883	4.328	28.906	1.00	22	below bound: anchor_p99, angle_p99
jolt	4	3	2.280	2.274	2.469	2.715	2.831	2.276–2.282	2.023	1.565	26.810	3.91	22	below bound: anchor_p99, angle_p99
rapier-par	1	3	8.050	7.957	9.652	9.842	9.925	8.016–8.075	8.019	21.434	31.160	1.00	42	below bound: anchor_p99, angle_p99
rapier-par	4	3	2.769	2.712	3.491	3.648	3.744	2.748–2.780	2.828	17.645	27.806	3.58	43	below bound: anchor_p99, angle_p99

Solver sweep

Four threads, the same iteration count in every engine (Jolt: velocity steps with 2 position steps; Rapier: solver iterations; Avian: substeps). oxijolt has no row at 2 velocity steps on joint_revolute: the run failed (below). The validation of many_pyramids here measured the height against tick 60, before the reference moved to tick 300 (results' README).

many_pyramids

variant	threads	iterations	warm mean ms	warm p99	anchor p99 m	angle p99 rad	height ratio	quality
avian-par	4	2	21.239	24.741	n/a	n/a	1.001619	below bound: final_speed_p99
avian-par	4	4	26.817	30.705	n/a	n/a	1.000369	within bounds
avian-par	4	8	38.792	43.791	n/a	n/a	1.000104	within bounds
jolt	4	2	7.282	14.078	n/a	n/a	1.000731	below bound: ground_penetration, final_speed_p99
jolt	4	4	6.017	6.561	n/a	n/a	0.999710	below bound: final_speed_p99
jolt	4	8	7.617	7.722	n/a	n/a	0.999942	within bounds
rapier-par	4	2	6.971	7.480	n/a	n/a	1.000737	below bound: final_speed_p99
rapier-par	4	4	8.386	9.109	n/a	n/a	0.999990	within bounds
rapier-par	4	8	10.768	11.240	n/a	n/a	0.999989	within bounds

joint_revolute

variant	threads	iterations	warm mean ms	warm p99	anchor p99 m	angle p99 rad	height ratio	quality
avian-par	4	2	6.537	7.126	17.074408	2.864168	n/a	below bound: anchor_p99, angle_p99
avian-par	4	4	9.834	10.811	21.198238	3.007380	n/a	below bound: anchor_p99, angle_p99
avian-par	4	8	16.063	17.121	22.885859	3.010580	n/a	below bound: anchor_p99, angle_p99
jolt	4	4	1.910	2.331	1.626429	0.493550	n/a	below bound: anchor_p99, angle_p99
jolt	4	8	2.143	2.634	1.065376	0.375391	n/a	below bound: anchor_p99, angle_p99
rapier-par	4	2	1.624	2.570	0.691739	0.311043	n/a	below bound: anchor_p99, angle_p99
rapier-par	4	4	2.340	3.234	0.203237	0.109502	n/a	below bound: anchor_p99, angle_p99
rapier-par	4	8	3.718	4.529	0.068752	0.040086	n/a	within bounds

Avian's physics schedule inside the update

Split runs with Avian's total step timer, their three repeats pooled like the rows above: the update as in the tables above, the time of Avian's physics schedule inside it, and the rest (Bevy's schedule, time and transform propagation, Avian's systems outside its physics schedule). The last column is the plain timing run's update. The first column minus the last is the timer's own cost plus the variation between runs; it is negative in four rows, down to -1.6 ms on keva.

scene	profile	threads	update mean ms	physics schedule mean ms	rest of the update mean ms	timing-run update mean ms
balls	matched	1	9.186	8.109	1.076	9.139
balls	matched	4	4.287	3.665	0.622	4.278
boxes	matched	1	11.416	11.052	0.364	11.396
boxes	matched	4	6.436	6.196	0.240	6.409
capsules	matched	1	10.101	9.389	0.712	10.119
capsules	matched	4	5.528	5.155	0.373	5.507
joint_ball	matched	1	21.160	19.537	1.623	21.046
joint_ball	matched	4	15.436	14.203	1.233	15.209
joint_fixed	matched	1	15.541	13.937	1.603	15.449
joint_fixed	matched	4	13.869	12.657	1.212	13.843
joint_prismatic	matched	1	50.676	48.123	2.552	50.235
joint_prismatic	matched	4	31.889	29.930	1.960	31.798
joint_revolute	matched	1	13.485	12.208	1.276	13.470
joint_revolute	matched	4	9.843	8.969	0.874	9.844
keva	matched	1	504.754	500.444	4.311	506.118
keva	matched	4	193.663	189.483	4.180	195.277
many_pyramids	matched	1	69.164	67.959	1.205	69.058
many_pyramids	matched	4	27.245	26.392	0.854	27.086
pyramid	matched	1	589.255	584.637	4.617	589.059
pyramid	matched	4	232.916	229.199	3.717	231.822

Quality at four threads

The validation runs at 1, 4, 8 and 16 threads gave the same digests, so the same measures, for every engine and scene (next section); the four-thread rows stand for all four.

variant	scene	profile	threads	iterations	height_ratio	ground_penetration	ground_penetration_max	fallen	final_speed_p99	ball_overlap	ball_overlap_max	anchor_p99	angle_p99	limit_p99	awake_at_end	violations
jolt	balls	matched	4	default	n/a	n/a	n/a	n/a	n/a	0.029945	0.262690	n/a	n/a	n/a	7600	none
jolt-4	balls	matched	4	4	n/a	n/a	n/a	n/a	n/a	0.030387	0.340679	n/a	n/a	n/a	7600	none
rapier-par	balls	matched	4	default	n/a	n/a	n/a	n/a	n/a	0.009950	0.301052	n/a	n/a	n/a	7600	none
avian-par	balls	matched	4	default	n/a	n/a	n/a	n/a	n/a	0.004417	0.287813	n/a	n/a	n/a	7600	none
jolt	balls	defaults	4	default	n/a	n/a	n/a	n/a	n/a	0.041929	0.295475	n/a	n/a	n/a	3743	none
jolt-4	balls	defaults	4	4	n/a	n/a	n/a	n/a	n/a	0.043679	0.334797	n/a	n/a	n/a	0	none
rapier-par	balls	defaults	4	default	n/a	n/a	n/a	n/a	n/a	0.009990	0.301052	n/a	n/a	n/a	0	none
avian-par	balls	defaults	4	default	n/a	n/a	n/a	n/a	n/a	0.004442	0.088102	n/a	n/a	n/a	0	none
jolt	capsules	matched	4	default	n/a	172.175339	172.175339	12	8.562654	n/a	n/a	n/a	n/a	n/a	3008	ground_penetration,fallen
jolt-4	capsules	matched	4	4	n/a	217.529099	217.529099	240	57.326023	n/a	n/a	n/a	n/a	n/a	3008	ground_penetration,fallen
rapier-par	capsules	matched	4	default	n/a	0.030036	0.210192	0	7.100881	n/a	n/a	n/a	n/a	n/a	3008	none
avian-par	capsules	matched	4	default	n/a	0.035773	0.266310	0	7.593885	n/a	n/a	n/a	n/a	n/a	3008	none
jolt	capsules	defaults	4	default	n/a	166.532562	166.532562	24	10.718449	n/a	n/a	n/a	n/a	n/a	3008	ground_penetration,fallen
jolt-4	capsules	defaults	4	4	n/a	244.233658	244.233658	198	49.629009	n/a	n/a	n/a	n/a	n/a	3008	ground_penetration,fallen
rapier-par	capsules	defaults	4	default	n/a	0.015109	0.210192	0	8.287679	n/a	n/a	n/a	n/a	n/a	3008	none
avian-par	capsules	defaults	4	default	n/a	0.012632	0.143123	0	7.562224	n/a	n/a	n/a	n/a	n/a	3004	none
jolt	joint_ball	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.075314	n/a	n/a	9974	none
jolt-4	joint_ball	matched	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.141170	n/a	n/a	9974	anchor_p99
rapier-par	joint_ball	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.058051	n/a	n/a	9974	none
avian-par	joint_ball	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	1.141398	n/a	n/a	9974	anchor_p99
jolt	joint_ball	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.069071	n/a	n/a	9974	none
jolt-4	joint_ball	defaults	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.132719	n/a	n/a	9974	anchor_p99
rapier-par	joint_ball	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.058051	n/a	n/a	9974	none
avian-par	joint_ball	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.140782	n/a	n/a	9974	anchor_p99
jolt	joint_fixed	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.008434	0.044251	n/a	11500	none
jolt-4	joint_fixed	matched	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.024531	0.108449	n/a	11500	angle_p99
rapier-par	joint_fixed	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.013977	0.032331	n/a	11500	none
avian-par	joint_fixed	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.035137	0.058193	n/a	11500	none
jolt	joint_fixed	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.008299	0.044173	n/a	11500	none
jolt-4	joint_fixed	defaults	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.023685	0.108335	n/a	11500	angle_p99
rapier-par	joint_fixed	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.014179	0.032753	n/a	0	none
avian-par	joint_fixed	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.015305	0.025398	n/a	11500	none
jolt	joint_prismatic	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.057239	0.044255	0.011519	16000	none
jolt-4	joint_prismatic	matched	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.107055	0.092380	0.021988	16000	anchor_p99
rapier-par	joint_prismatic	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.039868	0.039926	0.002505	16000	none
avian-par	joint_prismatic	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	2.770000	2.999875	1.551825	16000	anchor_p99,angle_p99,limit_p99
jolt	joint_prismatic	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.051587	0.039619	0.011875	16000	none
jolt-4	joint_prismatic	defaults	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.101996	0.092511	0.019823	16000	anchor_p99
rapier-par	joint_prismatic	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.039868	0.039926	0.002505	16000	none
avian-par	joint_prismatic	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	1.167864	2.699181	0.499149	16000	anchor_p99,angle_p99,limit_p99
jolt	joint_revolute	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.919359	0.346165	n/a	8000	anchor_p99,angle_p99
jolt-4	joint_revolute	matched	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	1.626429	0.493550	n/a	8000	anchor_p99,angle_p99
rapier-par	joint_revolute	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.203237	0.109502	n/a	8000	anchor_p99,angle_p99
avian-par	joint_revolute	matched	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	21.198238	3.007380	n/a	8000	anchor_p99,angle_p99
jolt	joint_revolute	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.906718	0.339715	n/a	8000	anchor_p99,angle_p99
jolt-4	joint_revolute	defaults	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	1.737172	0.493844	n/a	8000	anchor_p99,angle_p99
rapier-par	joint_revolute	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.203237	0.109502	n/a	8000	anchor_p99,angle_p99
avian-par	joint_revolute	defaults	4	default	n/a	n/a	n/a	n/a	n/a	n/a	n/a	12.288465	3.006289	n/a	8000	anchor_p99,angle_p99
jolt	many_pyramids	matched	4	2	1.000731	0.343223	0.394575	0	9.010623	n/a	n/a	n/a	n/a	n/a	8400	ground_penetration,final_speed_p99
jolt	many_pyramids	matched	4	4	0.999710	0.020040	0.021105	0	0.861535	n/a	n/a	n/a	n/a	n/a	8400	final_speed_p99
jolt	many_pyramids	matched	4	8	0.999942	0.007690	0.010547	0	0.010480	n/a	n/a	n/a	n/a	n/a	8400	none
rapier-par	many_pyramids	matched	4	2	1.000737	0.001469	0.002442	0	0.627694	n/a	n/a	n/a	n/a	n/a	8400	final_speed_p99
rapier-par	many_pyramids	matched	4	4	0.999990	0.001008	0.001216	0	0.000145	n/a	n/a	n/a	n/a	n/a	8400	none
rapier-par	many_pyramids	matched	4	8	0.999989	0.001286	0.001295	0	0.000414	n/a	n/a	n/a	n/a	n/a	8400	none
avian-par	many_pyramids	matched	4	2	1.001619	0.030292	0.047729	0	2.436389	n/a	n/a	n/a	n/a	n/a	8400	final_speed_p99
avian-par	many_pyramids	matched	4	4	1.000369	0.008015	0.010451	0	0.096379	n/a	n/a	n/a	n/a	n/a	8400	none
avian-par	many_pyramids	matched	4	8	1.000104	0.002184	0.002583	0	0.028137	n/a	n/a	n/a	n/a	n/a	8400	none
jolt	joint_revolute	matched	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	1.626429	0.493550	n/a	8000	anchor_p99,angle_p99
jolt	joint_revolute	matched	4	8	n/a	n/a	n/a	n/a	n/a	n/a	n/a	1.065376	0.375391	n/a	8000	anchor_p99,angle_p99
rapier-par	joint_revolute	matched	4	2	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.691739	0.311043	n/a	8000	anchor_p99,angle_p99
rapier-par	joint_revolute	matched	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.203237	0.109502	n/a	8000	anchor_p99,angle_p99
rapier-par	joint_revolute	matched	4	8	n/a	n/a	n/a	n/a	n/a	n/a	n/a	0.068752	0.040086	n/a	8000	none
avian-par	joint_revolute	matched	4	2	n/a	n/a	n/a	n/a	n/a	n/a	n/a	17.074408	2.864168	n/a	8000	anchor_p99,angle_p99
avian-par	joint_revolute	matched	4	4	n/a	n/a	n/a	n/a	n/a	n/a	n/a	21.198238	3.007380	n/a	8000	anchor_p99,angle_p99
avian-par	joint_revolute	matched	4	8	n/a	n/a	n/a	n/a	n/a	n/a	n/a	22.885859	3.010580	n/a	8000	anchor_p99,angle_p99
jolt	boxes	matched	4	default	1.000000	0.008708	0.009564	0	0.000221	n/a	n/a	n/a	n/a	n/a	1000	none
jolt-4	boxes	matched	4	4	1.000001	0.020343	0.024173	0	0.001542	n/a	n/a	n/a	n/a	n/a	1000	none
rapier-par	boxes	matched	4	default	1.000000	0.003215	0.003467	0	0.000173	n/a	n/a	n/a	n/a	n/a	1000	none
avian-par	boxes	matched	4	default	0.999981	0.047923	0.270944	0	10.806533	n/a	n/a	n/a	n/a	n/a	1000	final_speed_p99
jolt	boxes	defaults	4	default	1.000000	0.020000	0.259614	0	0.000000	n/a	n/a	n/a	n/a	n/a	0	none
jolt-4	boxes	defaults	4	4	1.000000	0.020001	0.258268	0	0.000000	n/a	n/a	n/a	n/a	n/a	0	none
rapier-par	boxes	defaults	4	default	1.000000	0.003210	0.003467	0	0.000000	n/a	n/a	n/a	n/a	n/a	0	none
avian-par	boxes	defaults	4	default	1.000398	0.007493	0.015598	0	0.057631	n/a	n/a	n/a	n/a	n/a	1000	none
jolt	pyramid	matched	4	default	1.000075	0.028015	0.034396	0	0.041453	n/a	n/a	n/a	n/a	n/a	42925	none
jolt-4	pyramid	matched	4	4	0.999857	0.020174	0.105019	0	0.181281	n/a	n/a	n/a	n/a	n/a	42925	final_speed_p99
rapier-par	pyramid	matched	4	default	0.994331	0.474314	0.601858	0	0.010267	n/a	n/a	n/a	n/a	n/a	42925	ground_penetration
avian-par	pyramid	matched	4	default	0.978940	0.333171	0.443739	0	0.188523	n/a	n/a	n/a	n/a	n/a	42925	ground_penetration,final_speed_p99
jolt	pyramid	defaults	4	default	0.999956	0.023687	0.043731	0	0.015515	n/a	n/a	n/a	n/a	n/a	42925	none
jolt-4	pyramid	defaults	4	4	1.000260	0.020040	0.079906	0	0.090711	n/a	n/a	n/a	n/a	n/a	42925	none
rapier-par	pyramid	defaults	4	default	0.994331	0.003792	0.043466	0	0.010267	n/a	n/a	n/a	n/a	n/a	42924	none
avian-par	pyramid	defaults	4	default	1.000186	0.020856	0.033686	0	0.107847	n/a	n/a	n/a	n/a	n/a	42925	final_speed_p99
jolt	many_pyramids	matched	4	default	1.000000	0.005366	0.008088	0	0.026427	n/a	n/a	n/a	n/a	n/a	8400	none
jolt-4	many_pyramids	matched	4	4	0.999941	0.020040	0.021105	0	0.861535	n/a	n/a	n/a	n/a	n/a	8400	final_speed_p99
rapier-par	many_pyramids	matched	4	default	1.000000	0.001008	0.001216	0	0.000145	n/a	n/a	n/a	n/a	n/a	8400	none
avian-par	many_pyramids	matched	4	default	1.000166	0.008015	0.010451	0	0.096379	n/a	n/a	n/a	n/a	n/a	8400	none
jolt	many_pyramids	defaults	4	default	1.000000	0.005852	0.008047	0	0.000000	n/a	n/a	n/a	n/a	n/a	0	none
jolt-4	many_pyramids	defaults	4	4	1.000000	0.020098	0.021347	0	1.563359	n/a	n/a	n/a	n/a	n/a	6300	final_speed_p99
rapier-par	many_pyramids	defaults	4	default	1.000000	0.001001	0.001216	0	0.000000	n/a	n/a	n/a	n/a	n/a	0	none
avian-par	many_pyramids	defaults	4	default	1.000009	0.003246	0.004294	0	0.047213	n/a	n/a	n/a	n/a	n/a	7140	none
jolt	keva	matched	4	default	0.246323	68.152679	68.152679	13	13.694255	n/a	n/a	n/a	n/a	n/a	38270	height_ratio,ground_penetration,fallen,final_speed_p99
jolt-4	keva	matched	4	4	0.180675	433.153961	433.153961	2269	54.213150	n/a	n/a	n/a	n/a	n/a	38270	height_ratio,ground_penetration,fallen,final_speed_p99
rapier-par	keva	matched	4	default	1.000506	0.010382	0.017400	0	0.269966	n/a	n/a	n/a	n/a	n/a	38270	final_speed_p99
avian-par	keva	matched	4	default	0.228516	361.583618	361.583618	779	30.800726	n/a	n/a	n/a	n/a	n/a	38270	height_ratio,ground_penetration,fallen,final_speed_p99
jolt	keva	defaults	4	default	0.241268	109.771332	109.771332	3793	27.464169	n/a	n/a	n/a	n/a	n/a	38267	height_ratio,ground_penetration,fallen,final_speed_p99
jolt-4	keva	defaults	4	4	0.149925	380.599731	380.599731	4602	44.016476	n/a	n/a	n/a	n/a	n/a	38270	height_ratio,ground_penetration,fallen,final_speed_p99
rapier-par	keva	defaults	4	default	1.000506	0.010382	0.017400	0	0.269966	n/a	n/a	n/a	n/a	n/a	38270	final_speed_p99
avian-par	keva	defaults	4	default	0.234649	17.720209	17.720209	12	13.007080	n/a	n/a	n/a	n/a	n/a	38263	height_ratio,ground_penetration,fallen,final_speed_p99

What the bounds caught, at both profiles unless a profile is named:

• capsules: oxijolt's heap pushes capsules through the 20 cm ground box, 12 of 3008 at matched and 24 at defaults (240 and 198 with 4 velocity steps). Rapier and Avian lose none. The scene runs without continuous collision in oxijolt at both profiles (Jolt's default motion quality is discrete).
• keva: oxijolt's and Avian's towers collapse (a quarter of the tick-300 height is left, and bodies fall off the ground); Rapier's stand, still moving at 0.27 m/s (99th percentile) at tick 600.
• pyramid: oxijolt keeps the pyramid and its lowest boxes 3 cm in the ground. At matched, Rapier's deepest box ends 47 cm in the ground and Avian's 33 cm; at defaults (with each engine's own contact settings) they stay under 5 cm.
• boxes: Avian's stack at matched still moves at 10.8 m/s (99th percentile) at tick 600.
• joint_ball, joint_prismatic, joint_revolute: Avian's joints come apart (anchor errors at matched of 1.1 m, 2.8 m and 21 m); oxijolt and Rapier stay within 10 cm on the first two.
• joint_revolute is outside the bounds in every engine: anchor error 0.20 m for Rapier, 0.92 m for oxijolt. Rapier needs 8 iterations to stay within them (solver sweep).
• balls: every engine ends within the 5 cm overlap; during the impacts at matched the deepest overlap is 26 to 30 cm in every engine.
• With 4 velocity steps, oxijolt leaves the bounds on joint_ball, joint_fixed and joint_prismatic (joint error) and on pyramid at matched and many_pyramids (final speed), where 10 steps stay within them.

Determinism across thread counts

All 80 groups (ten scenes, two profiles, jolt, jolt-4, rapier-par and avian-par) gave the same digest at 1, 4, 8 and 16 threads on all 600 ticks, and every group moved, so the check is not empty. The table is in summary.md. This is one binary per engine on one machine; it says nothing about other machines or compilers.

Features

Checked on 2026-10-07 against oxijolt at this commit, rapier3d 0.36.0 (tag v0.36.0) and avian3d 0.7.0 (tag v0.7.0), in each crate's source and on crates.io. "None found" names where we looked.

	oxijolt	Rapier 0.36.0	Avian 0.7.0
Character controller	Jolt's CharacterVirtual, stairs and floor sticking (features)	KinematicCharacterController (source)	MoveAndSlide utilities for writing one, not a controller (source)
Vehicles	wheeled, tracked and motorcycle, with engine, gearbox and differentials (guide)	ray-cast vehicle DynamicRayCastVehicleController (source)	none found in src/ (tree)
Ragdolls	a ragdoll type from a skeleton, with motors and a skeleton mapper (guide)	none built in; built from joints (joints)	none built in; built from joints (joints)
Soft bodies	cloth, pressure and volume bodies (guide)	yes, new in 0.36.0 (changelog, SoftBodySet)	not included ("Elasticity (soft bodies)" is listed as not covered, source)
Joint kinds	12: fixed, point, distance, hinge, slider, cone, swing-twist, six-DOF, gear, rack and pinion, pulley, path (guide)	8 impulse-joint builders (fixed, spherical, revolute, prismatic, rope, spring, pin-slot, generic) and multibody joints (source)	5: fixed, distance, prismatic, revolute, spherical (source)
Contact events	added, persisted and removed contacts, a contact listener (guide)	EventHandler, collision and contact-force events (source)	CollisionStart and CollisionEnd events (source)
Same results for any thread count	tested with 1 and 4 workers (determinism); measured here at 1, 4, 8 and 16	promised for any Rayon pool size and tested (test); measured here	no promise found in the docs; the narrow phase keeps a serial loop "to preserve determinism" (source); measured here
Same results across operating systems	opt-in feature cross-platform-deterministic; CI compares a wasm build with native Linux for one scene (determinism)	enhanced-determinism feature, with a snapshot test run on native and wasm32-wasip1 (test)	enhanced-determinism feature (source); its cross-platform test is 2D (test)
Save and restore	the same world's state, for rollback and replay (guide)	serde snapshots of the whole world (serde-serialize, manifest)	none built in; serialize derives serde for components (manifest)
Double precision	f64 world positions (double-precision, features)	the rapier3d-f64 crate (manifest)	f64 feature (manifest)
WASM	oxijolt builds for wasm32-unknown-unknown from a release archive or from source, single-threaded, tested in CI in Node and headless Chrome (building); Jolt itself builds for WASM (JoltPhysics.js)	yes, wasm32-wasip1 in its CI test (test) and JavaScript bindings (bindings)	through Bevy, which supports WASM (Bevy); no WASM job in Avian's CI (ci.yml)
Language	Rust over C++ (Jolt and joltc) (lineage)	pure Rust (manifest)	pure Rust (manifest)
Build requirements	on x86_64 Windows and Linux and on Apple silicon a verified prebuilt download; elsewhere a C++ toolchain and CMake 3.20 (building)	Rust only (manifest)	Rust only (manifest)
Needs Bevy	no (manifest)	no; bevy_rapier is separate (bindings)	yes (manifest)
Minimum Rust	1.88 (manifest)	1.86 (manifest)	not declared; its Bevy 0.19.1 (the version this comparison's lockfile resolves) needs 1.95 (Bevy manifest)
Licence	MIT or Apache-2.0; Jolt and joltc MIT (lineage)	Apache-2.0 (manifest)	MIT or Apache-2.0 (manifest)

Where oxijolt loses

• A C++ toolchain and CMake outside x86_64 Windows and Linux, Apple silicon and wasm32-unknown-unknown, and for any build no release archive fits: oxijolt then compiles Jolt and joltc with MSVC, GCC or Clang through CMake (building). Rapier and Avian are pure Rust.
• WASM is single-threaded. oxijolt builds for wasm32-unknown-unknown without threads; CI runs it in Node and headless Chrome (building). Rapier runs a WASM test in CI and has JavaScript bindings.
• macOS comes prebuilt and is tested in CI on Apple silicon only; Intel Macs build from source. This comparison runs on Windows and Linux only: its harness reads CPU time and memory from those two systems.
• Same results across operating systems are an opt-in feature; CI checks wasm in Node and Chrome against Linux native, but not native operating systems against each other yet. Rapier tests a snapshot on native and WASM targets.
• Save and restore covers the same world with an unchanged structure only (state).
• Speed on scenes of many contacts. At matched settings Rapier's step is 1.9 to 2.2 times faster on boxes, 1.3 to 1.85 times on pyramid, 1.2 to 1.75 times on capsules, 1.4 times on balls up to 8 threads, and 1.6 times on keva at one thread. Jolt's default of 10 velocity steps is part of it; with 4 Rapier is still 1.2 to 1.65 times faster on boxes.
• joint_ball at 8 and 16 threads. oxijolt's warm step is 10.7 ms at 8 threads against 4.5 ms at 4, with per-run means from 8.3 to 12.1 ms, while its first 60 ticks keep scaling (3.8 ms). Rapier is 3.4 to 4.3 times faster there. We have not found the cause.
• Quality on three scenes. Capsules through the ground, collapsing keva towers, and on joint_revolute a joint error 4.5 times Rapier's (above).
• Building worlds with many joints. World creation plus the first tick at four threads takes oxijolt 225 ms on joint_ball, 235 ms on joint_fixed and 142 ms on joint_prismatic, against Rapier's 43, 54 and 62 ms; oxijolt's first tick alone is under 6 ms there.
• Sleeping at defaults. With each engine's own thresholds Rapier is 5 times faster on balls and 8 to 9 times on boxes.
• Two velocity steps. The solver sweep's joint_revolute at 2 velocity steps fails in oxijolt: at tick 339 the step drops contacts because its body-pair buffer, sized at 10 steps, is full. The run is in failures.tsv and has no row.

In a real game

The owner's numbers from the game VOIDRUN, which replaced Rapier's kinematic character controller with oxijolt's CharacterVirtual (the table as given). The game rows are a release build on a weak 4-core machine. They are the game's own observations, not runs of this repository, and the two sides did not always time the same work:

What	Rapier	Jolt	What each side timed
Whole walking system, 60 NPCs in the near band, mean per tick	3.43 ms	0.83 ms	the game's walking system with each engine; a system-level observation, not an engine speed
One controller move in the game	86-156 us	16-19 us p50 / 28-43 us p99	Rapier: in the game, statistic and window not recorded; Jolt: in the game, with the game's wrapper. No ratio.
One controller move, the binding alone	n/a	2.7 us p50 / 6 us p99	update_character with terrain in the binding's own benchmark: 30 characters on a flat 3 x 3 chunk scene, on an Intel Core i9-11900K under Windows 11, not the game's machine and not the machine of the runs above (benchmarks); no Rapier counterpart
Physics world step p50/p99	110 / 268 us	45 / 111 us	the world step; the setups differ in damping and sleeping, and Rapier's item gravity was applied before its timer, Jolt's inside it
Refresh before queries when the world changed, p99	3.8 ms	164 us	Rapier detect_collisions (broad and narrow phase) against Jolt optimize_broad_phase: different work, no ratio

The game ran rapier3d 0.36.0: its manifest required 0.36 with no patch, and 0.36.0 is the only 0.36 release. Rapier 0.36 already serves queries from its broad phase's tree, so the refresh row is not a query-structure rebuild; the game's refresh ran collision detection. The owner's caveats: the gains for the whole game are smaller than the bare controller's, because the rest of the frame is unchanged, and the game's own wrapper around the controller dominates its cost in the game.

A head-to-head of the character controllers on one scene is not part of this comparison.

Reproduce

On Linux with a C++ toolchain, CMake 3.20 or newer, Python 3 and Rust 1.95 or newer (Bevy 0.19.1, which Avian pulls in, needs it; oxijolt alone needs 1.88), from the repository root, one cargo process at a time on an otherwise idle machine:

python3 benches/comparison/scripts/build_variants.py          # six binaries in target/comparison-bins
R=benches/comparison/results/my-run
python3 benches/comparison/scripts/campaign.py $R validate
python3 benches/comparison/scripts/campaign.py $R sweep-validate
for rep in 1 2 3; do
  for plan in main extras split sweep; do
    python3 benches/comparison/scripts/campaign.py $R $plan --repeat $rep
  done
done
python3 benches/comparison/scripts/build_times.py $R
python3 benches/comparison/scripts/make_doc.py $R              # copies the tables into this page

The same scripts run on Windows (load from load.ps1, no pinning). comparison help lists the options of the harness itself.