Rendering and AI Optimization in RTS games: Enhancing performance with Entity Component Systems in Unreal Engines 5

Abstract

This thesis presents a game prototype made with Unreal Engine 5’s Mass framework.
The point is to see how well a data-oriented Entity Component System (ECS) can handle
complex logic and rendering needs of RTS games with over 50000 units. The prototype
uses the mass plugins fragments for navigation targets, avoiding, sensing, combat, and
showing things visually along with processors that apply the logic via queries. The
thesis examines two ways to render: one used Instanced Static Meshes with Vertex
Animation, and the other used Niagara mesh renderer with Vertex Animation. After
implementing the prototype, an evaluation framework around the prototype was built so
that performance measurements that could be repeated could be taken. The assessment
concentrated on three main questions for research: (1) How well do pathfinding and unit
movement hold up as unit counts and environment complexity grow? (2) How well do
the two rendering paths (ISM-based and Niagara-based) scale in practice? (3) How can
simplified interactions (collisions, melee combat) be implemented efficiently at scale?
The results show that the Mass framework can handle RTS gameplay with up to
50,000 units while still being stable at 60 FPS. Pathfinding and movement work well in
simple settings, but they become much more expensive in more complicated ones. The
ISM-based renderer works well with a moderate number of units (up to about 10,000),
but it has trouble with GPU times when there are more units. The renderer that works
better with a lot of units (up to 50,000 and beyond) is the one based in Niagara. It also
has more consistent performance. Even in crowded situations, spatial partitioning and
time-sliced collision checks can handle a lot of entities without putting too much strain
on the CPU. This thesis demonstrates that Unreal’s Mass framework provides a robust
foundation for real-time strategy games.

Date of Award	2025
Original language	English (American)
Supervisor	Martin Kocur (Supervisor)