Real-Time GI vs Probe-Based Lighting Performance

Lighting choices in game development represent a fundamental trade-off between computational cost, visual fidelity, and development workflow. Real-time global illumination (RTGI) calculates dynamic light bounces per frame, delivering realistic, fully reactive lighting at the expense of heavy GPU compute and memory bandwidth. In contrast, probe-based lighting relies on pre-computed or dynamically sampled grids of light probes that store irradiance data, offering high frame-rate predictability and minimal runtime computation at the expense of static scene constraints and increased memory footprints.

GPU Compute and Frame Times

Real-time global illumination relies on ray tracing (hardware-accelerated or software-based) or screen-space approximations to calculate indirect light bounces. This introduces significant per-frame GPU workloads:

  • Ray Tracing Overheads: Traversing Bounding Volume Hierarchies (BVH) and tracing rays to evaluate intersection points demands massive compute throughput, heavily reducing base frame rates on low-to-mid-tier GPUs.
  • Denoising Passes: Because full real-time path tracing is too expensive, RTGI uses low ray counts per pixel. This requires complex spatial and temporal denoising filters to reconstruct a clean image, adding several milliseconds to the render pipeline.

Probe-based lighting shifts this computation almost entirely out of the per-frame render loop. Runtime shaders only evaluate low-order Spherical Harmonics (SH) per vertex or pixel by interpolating nearby probes. This lookup requires minimal arithmetic logic unit (ALU) operations, allowing games to maintain high, consistent frame rates even on mobile devices and older consoles.

Memory and Storage Footprints

The architectural differences between both approaches create opposing demands on system memory:

  • RTGI Memory Demands: Dynamic GI requires dedicated VRAM for acceleration structures, G-buffers, surface caches, and temporal history buffers. While it saves disk storage by eliminating baked lightmaps, its runtime VRAM consumption scales aggressively with rendering resolution.
  • Probe-Based Storage Demands: Probes require little VRAM for active execution, but large levels with high probe density require significant disk space and RAM to store the baked irradiance coefficients. Streaming high-density probe data across large open worlds can introduce streaming bandwidth bottlenecks.

CPU Overhead and Geometry Updates

RTGI requires the CPU to manage and dispatch dynamic scene changes continuously:

  • BVH Management: Dynamic objects, such as skeletal meshes and destructible props, must have their local BVH rebuilt or refitted by the CPU and GPU every frame to register accurate light bounces.
  • Draw Call Management: Probe systems bypass runtime geometry acceleration entirely, keeping the CPU render thread unencumbered by lighting updates.

Scalability and Hardware Targeting

Deploying games across multiple platforms highlights the performance disparity between these two systems:

  • RTGI Scalability: Dynamic GI features—such as Unreal Engine's Lumen or custom ray-traced solutions—generally require modern target platforms (PC with dedicated ray-tracing hardware, PlayStation 5, Xbox Series X/S). Scaling down to lower-end hardware often requires dropping RTGI entirely in favor of screen-space approximations or fallback systems.
  • Probe Scalability: Probe grids function uniformly across hardware generations. Performance is decoupled from scene complexity, making probe-based methods the industry standard for competitive titles, mobile games, and high-framerate VR applications where performance budgets are non-negotiable.