21 posts tagged with "release"

View All Tags

v0.35 - Rapier vs Box2D vs Jolt vs Box3D vs Godot Physics

August 8, 2026 · 5 min read

Apps in a Cup

Godot Rapier now runs on Rapier 0.35.

Both 2D and 3D builds now always have parallel SIMD solver and cross platform determinism enabled.

Here is the repo used to run the tests benchmarks-repo.

2D: Rapier vs Box2D

Rapier is the only engine that keeps every scene stable: Box2D collapses the pyramid and blows the joints apart, those are marked as unstable.

Issue on godot-box2d-v3 side: godot-box2d-v3/issues/9 godot-box2d-v3/issues/8

Engine	Version	Addon
Rapier2D	v0.35	godot-rapier2d
Box2D	v3.1	godot-box2d-v3
Godot Physics 2D	v4.7	ships with Godot

Rapier2D	Box2D	Godot Physics
20.0 ms 1.8 cores	21.3 ms 3.8 cores unstable	147 ms 1.2 cores
11.9 ms 2.1 cores	11.6 ms 3.8 cores	55.1 ms 1.3 cores
9.3 ms 1.7 cores	6.7 ms 2.9 cores	71.1 ms 1.3 cores
6.2 ms 1.2 cores	4.6 ms 1.6 cores	57.2 ms 1.1 cores
10.2 ms 1.5 cores	4.9 ms 2.6 cores unstable	unstable never finishes

Benchmark	1st	2nd	3rd
Pyramid	Rapier2D	Box2D 1.07x	Godot Physics 7.35x
Smash	Box2D	Rapier2D 1.02x	Godot Physics 4.76x
Mixed shape pile	Box2D	Rapier2D 1.38x	Godot Physics 10.60x
Query storm	Box2D	Rapier2D 1.37x	Godot Physics 12.56x
Joint grid	Box2D	Rapier2D 2.06x	Godot Physics never finishes

Unstable runs, whose times are not comparable: Box2D on the pyramid (collapses) and on the joint grid (infinite displacement). Godot Physics does not finish the joint grid within 60 s.

3D: Rapier vs Jolt vs Box3D

Rapier3D is fastest on queries and joints and stays close on the rest; Jolt crashes the joint grid. Box3D became multi-threaded in godot-box3d v0.2.4 and is now the fastest on three scenes — the numbers below were re-measured against it.

Engine	Version	Addon
Rapier3D	v0.35	godot-rapier3d
Jolt	v5.5.0	ships with Godot v4.7
Box3D	v0.1	godot-box3d v0.2.4
Godot Physics 3D	v4.7	ships with Godot

Rapier3D	Jolt	Box3D	Godot Physics
5.9 ms 1.0 cores	26.4 ms 1.1 cores	6.7 ms 1.1 cores	73.3 ms 1.0 cores
14.6 ms 2.8 cores	26.9 ms 4.7 cores	13.5 ms 2.7 cores	48.0 ms 1.4 cores
11.3 ms 2.1 cores	13.7 ms 3.6 cores	9.7 ms 2.4 cores	84.9 ms 1.7 cores
8.9 ms 1.7 cores	unstable never finishes	10.3 ms 1.8 cores unstable	26.1 ms 1.2 cores
19.1 ms 2.8 cores	12.1 ms 5.0 cores	9.6 ms 2.9 cores	100 ms 1.3 cores

Benchmark	1st	2nd	3rd	4th
Query storm	Rapier3D	Box3D 1.12x	Jolt 4.46x	Godot Physics 12.35x
Pyramid	Box3D	Rapier3D 1.08x	Jolt 1.99x	Godot Physics 3.55x
Mixed shape pile	Box3D	Rapier3D 1.16x	Jolt 1.41x	Godot Physics 8.73x
Joint grid	Rapier3D	Box3D 1.15x	Godot Physics 2.92x	Jolt never finishes
Smash	Box3D	Jolt 1.26x	Rapier3D 1.99x	Godot Physics 10.40x

Unstable runs, whose times are not comparable: Box3D on the joint grid (infinite displacement). Jolt does not finish the joint grid within 60 s, and Godot Physics does not finish the drop scene.

experimental-threads is no longer enabled

Official builds used to enable experimental-threads so the physics server could be driven from other threads. It cost 7-51% depending on the scene, and it turned out to be unnecessary: Godot already serialises every call onto a single physics thread through PhysicsServer3DWrapMT and its command queue, so the extra synchronisation was protecting a call path that is never concurrent. Godot Physics, Jolt and Box3D all take the same no-locking approach.

New features

Rope and fixed joints, in 2D and 3D. Godot has no equivalent of either. Reference.
Per-body solver iterations: stabilize one important body without paying for it globally. Reference.
Per-joint stiffness and toggling: joint_set_softness, joint_set_enabled for breakable constraints. Reference.
Joint warm starting: stiff joint assemblies converge much better, at no measurable cost.
New project settings for motion recovery, shape casting and query behaviour.