This repository contains a collection of benchmarks for popular Entity-Component-System (ECS) frameworks. The benchmarks cover different aspects of ECS frameworks, such as update systems, component additions/removals, and entity creation/destruction. It's important to note that different ECS frameworks have different strengths and weaknesses. For example, some frameworks might excel in adding/removing components, while others might be better at creating/destroying entities or have query support. Therefore, it's crucial to choose an ECS framework based on your specific requirements.

ECS (Entity-Component-System) Frameworks:

• EntityX
• EnTT
• Ginseng
• mustache
• flecs
• pico_ecs
• gaia-ecs

Main feature of an ECS is the iterating over a lot of entities and using system to update the components. The results of these benchmarks should be used as a starting point for your own benchmarking efforts.

Graph shows cost per entity, tables shows total cost. lower is faster.

While this benchmark only includes up to 6 components and 7 small systems, it's important to note that Entity-Component-Systems can become much more complex in the wild, with hundreds of components and systems. Therefore, it's crucial to always benchmark your specific cases and systems when necessary and compare results. Choose an ECS framework based on its features, for example, EnTT offers resource management and event handling, while flecs provides useful add-ons and querying, and EntityX includes a built-in world/system manager.

To evaluate a framework, look at the examples and API design, and pick the one that suits your needs.

All benchmarks are located in the benchmark/benchmarks/ directory and write with with the google/benchmark library. Each benchmark uses an example application for each framework (see src/ directory), and every example application has specific base features implemented (see src/base).

1. PositionComponent: includes x and y coordinates.
2. VelocityComponent: includes x and y coordinates for movement.
3. DataComponent: includes some arbitrary data.
4. HealthComponent: Hero/Monster data includes HP/MaxHP and status.
5. DamageComponent: Hero/Monster data includes damage.
6. SpriteComponent: Hero/Monster ASCII character as sprite.

1. MovementSystem: updates the PositionComponent with a constant VelocityComponent.
2. DataSystem: updates the DataComponent with arbitrary data.
3. MoreComplexSystem: updates components with random data and arbitrary information.
4. HealthSystem: update Hero/Monster health (update HP and status).
5. DamageSystem: update Hero/Monster health by taking damage.
6. SpriteSystem: update Hero/Monster ASCII character depending on health and type.
7. RenderSystem: "render"(write) Hero/Monster character sprite into a "frame buffer"(string buffer).

Benchmarks for more common features, such as "Creating entities", "Adding and removing components", and others.

• Entity Creation
• Entity Destruction
• Component Retrieval
• Adding and removing components
• Iterating entities/quires

• OS: Linux 64-Bit (Kernel: 6.10.4)
• CPU: 3.13GHz @ 12Cores
• RAM: 47GB
• Compiler: gcc (GCC) 14.2.1

Graph shows cost per entity, tables shows total cost. lower is faster.

Graph shows cost per entity, tables shows total cost. lower is faster.

Graph shows cost per entity, tables shows total cost. lower is faster.

Note:

• Get one non-const component
  1. PositionComponent

Graph shows cost per entity, tables shows total cost. lower is faster.

Note:

• Get non-const- and const-component
  1. PositionComponent
  2. const VelocityComponent

Graph shows cost per entity, tables shows total cost. lower is faster.

Note:

• Not every entity has three components, some has only two
• Get two non-const- and const-component(s)
  1. PositionComponent
  2. const VelocityComponent
  3. DataComponent (optional)

Graph shows cost per entity, tables shows total cost. lower is faster.

Note:

• Remove and add PositionComponent

Cost per entity	Cost of all entities

Tables shows total cost. lower is faster.

Note:

• Systems used
  1. MovementSystem
  2. DataSystem

Cost per entity	Cost of all entities

Tables shows total cost. lower is faster.

Note:

• Systems used
  1. MovementSystem
  2. DataSystem
• Not every entity has all three components, some got removed

Cost per entity	Cost of all entities

Tables shows total cost. lower is faster.

Note:

• Systems used
  1. MovementSystem
  2. DataSystem
  3. MoreComplexSystem
  4. HealthSystem
  5. DamageSystem
  6. SpriteSystem
  7. RenderSystem
• * EnTT iterate components via views
• ** EnTT iterate components via runtime views
• *** EnTT iterate components via groups:
  1. DataSystem: No Group, use registry.view<DataComponent>. (Can't group a single component)
  2. MovementSystem: Partial-owning group, registry.group<PositionComponent>(entt::get<const VelocityComponent>)
  3. MoreComplexSystem: Full-owning group, registry.group<PositionComponent, VelocityComponent, DataComponent>()
• **** EnTT iterate components via view and uses a stable component (StablePositionComponent)
• ***** gaia-ecs (SoA) iterate via SoA Layout using an Position- and Velocity-SoA-component

Cost per entity	Cost of all entities

Tables shows total cost. lower is faster.

Note:

• Systems used
  1. MovementSystem
  2. DataSystem
  3. MoreComplexSystem
  4. HealthSystem
  5. DamageSystem
  6. SpriteSystem
  7. RenderSystem
• Not every entity has all three components, some got removed
• * EnTT iterate components via views
• ** EnTT iterate components via runtime views
• *** EnTT iterate components via groups:
  1. DataSystem: No Group, use registry.view<DataComponent>. (Can't group a single component)
  2. MovementSystem: Partial-owning group, registry.group<PositionComponent>(entt::get<const VelocityComponent>)
  3. MoreComplexSystem: Full-owning group, registry.group<PositionComponent, VelocityComponent, DataComponent>()
• **** EnTT iterate components via view and uses a stable component (StablePositionComponent)
• ***** gaia-ecs (SoA) iterate via SoA Layout using an Position- und Velocity-SoA-component

If you have any improvements to the ECS-examples for any of the frameworks, feel free to make a pull request or open an issue. The example(s) for each framework can be found in src/, and benchmarks are located in benchmarks/benchmarks/ for more details.

Additionally, you can write tests for the framework example and add some metadata to the plot.config.json file. Any contributions are greatly appreciated!

TODO: make more detailed "how to add framework"

Read CONTRIBUTING.md for more details.

0. Install Taskfile
1. Run Build: task build
2. Run (All) Benchmarks: task benchmark
3. Plot Graphs: task plot:all

Or run task generate

You can now find the benchmark-results in reports/.

You need python (>=3.8.0) and some dependencies to run gen-benchmark-report (plotting), install pipx and poetry.

Then install the dependencies for the script:

cd ./scripts/gen-benchmark-report && poetry install

Or run the script directly:

task plot:all

task generate:readme

1. Configure and build benchmarks, see Building Details:
   1. cmake -S . -B ./build
   2. cmake --build ./build
2. Run benchmark(s), with run-benchmarks.sh or each:
   • ./build/benchmark/benchmarks/entt/ecs-benchmark-entt (console output) (optional)
   • ./build/benchmark/benchmarks/entt/ecs-benchmark-entt --benchmark_format=json > ./reports/entt.json (write json report)
3. Plot Reports, after generating json reports, see plot-results.sh:
   1. pipx run --spec ./scripts/gen-benchmark-report gen-benchmark-report -i ./info.json gen-plot ./reports/entityx.json ./reports/entt.json ./reports/ginseng.json ./reports/mustache.json ./reports/openecs.json ./reports/flecs.json (generate graphs)
   2. pipx run --spec ./scripts/gen-benchmark-report gen-benchmark-report -i ./info.json gen-results-md ./reports/entityx.json ./reports/entt.json ./reports/ginseng.json ./reports/mustache.json ./reports/openecs.json ./reports/flecs.json (generate full report)

cmake -G Ninja -S . -B build
cmake --build build --target ecs-benchmark-entt -j 4
./build/benchmark/benchmarks/entt/ecs-benchmark-entt

You can use -DCMAKE_BUILD_TYPE=Debug to enable Sanitizers.

• Dependency Setup
• Building Details
• Project Template, mix of: starter_template and ModernCppStarter
  • Forked cpp_vcpkg_project
• Google benchmark
• https://github.com/SanderMertens/ecs-faq
• https://github.com/jslee02/awesome-entity-component-system

Entity Component Systems (ECS) are a form of decomposition that completely decouples entity logic and data from the entity "objects" themselves. The Evolve your Hierarchy article provides a solid overview of EC systems and why you should use them.

Version: 1.1.2 (Apr 2023)

EnTT is a header-only, tiny and easy to use library for game programming and much more written in modern C++.

Version: v3.13.2

Ginseng is an entity-component-system (ECS) library designed for use in games.

The main advantage over similar libraries is that the component types do not need to be listed or registered. Component types are detected dynamically.

Any function-like object can be used as a system. The function's parameters are used to determine the required components.

Version: 1.1 (Dec 2021)

A fast, modern C++ Entity Component System

Version: 0.2 (Feb 2024)

Open Ecs is an Entity Component System that uses metaprogramming, cache coherency, and other useful tricks to maximize performance and configurability. It is written in c++11 without further dependencies.

Version: 0.1.101 (Apr 2017)

Flecs is a fast and lightweight Entity Component System that lets you build games and simulations with millions of entities.

Version: v4.0.1

A collection of cross-platform single header libraries written in C. Pure and simple ECS.

Version: 2.3 (Sep 2023)

Gaia-ECS is a fast and easy-to-use ECS framework.

Version: v0.8.6