VRAD-RTX

GPU-accelerated lightmap compiler for the Source Engine, built on NVIDIA OptiX and CUDA.

VRAD-RTX is a fork of Valve's Source SDK 2013 that replaces the performance-critical ray tracing stages of VRAD (the Source Engine's radiosity lighting tool) with hardware-accelerated equivalents using NVIDIA's OptiX 9.1 RT cores.

Features

• GPU Visibility Matrix — Accelerated VisMatrix generation via OptiX ray tracing
• GPU Shadow Ray Batching — Thread-local batched shadow rays for direct lighting, traced on the GPU
• GPU Radiosity Bounces — Light bounce calculations offloaded to CUDA with per-thread streams
• Hardware Profiling — Built-in system RAM, VRAM, CPU, and GPU usage logging
• Automated Regression Testing — PowerShell 7.6+ test scripts with bit-identical and visual parity checks against the original vrad.exe
• CPU Parity — CPU code path remains fully functional and produces identical results to stock VRAD

Performance

The GPU path accelerates the most expensive phases of lightmap compilation:

Phase	Speedup
Visibility Matrix	~6 s faster
Direct Lighting	~16 s faster
Radiosity Bounces	~29 s faster

Results measured on a map with 100 bounces; actual gains depend on map complexity and GPU hardware.

Requirements

• OS: Windows 10/11 (x64)
• CPU: x64 processor; AVX2/FMA3 recommended for -avx2 mode (Intel Haswell/2013+ or AMD Zen/2017+)
• GPU: NVIDIA GPU with RT cores (RTX 20-series or newer)
• CUDA Toolkit: 12.x
• OptiX SDK: 9.1
• Compiler: Visual Studio 2022 (MSVC v143, Windows SDK 10.0.22621.0+)
• Runtime: Source SDK Base 2013 Multiplayer (installed via Steam)
• Testing: PowerShell 7.6+, Python 3.13+

Building

1. Clone the repository:

   git clone https://github.com/legoj15/vrad-rtx.git
   

2. Navigate to src and generate the Visual Studio solution:

   createallprojects.bat

3. Open everything.sln in Visual Studio 2022 and build the vrad_dll project in Release | x64.

4. The built vrad_rtx.exe and vrad_dll.dll are copied to game/bin/x64/ automatically by the post-build step.

Usage

Run VRAD-RTX as a drop-in replacement for vrad.exe:

# CPU only (identical to stock VRAD)
vrad_rtx.exe path/to/mapname

# GPU accelerated
vrad_rtx.exe -cuda path/to/mapname

All standard VRAD command-line options are supported.

Testing

The test suite lives in game/bin/x64/ and validates correctness against the stock compiler:

# Quick regression test (reference → control → GPU, with visual diff)
.\test_vrad_optix_quick.ps1

# Full regression test
.\test_vrad_optix.ps1

Tests compare BSP lightmaps byte-for-byte against both the original vrad.exe and the CPU-only vrad_rtx.exe path, with a visual screenshot fallback using tgadiff.exe when binary comparison fails.

Project Structure

src/utils/vrad/              # Modified VRAD source
  vrad.cpp                   # Main entry point (timing, GPU flag)
  lightmap.cpp               # Core lightmap sampling & direct lighting
  vismat.cpp                 # Visibility matrix (GPU path)
  direct_lighting_gpu_vrad.cpp  # GPU direct lighting integration
  hardware_profiling.cpp     # System resource monitoring
  vrad_dll.vpc               # VPC project definition
game/bin/x64/                # Runtime & test harness
  test_vrad_optix_quick.ps1  # Quick regression test
  test_vrad_optix.ps1        # Full regression test
  bsp_diff_lightmaps.py      # Lightmap comparison utility
  tgadiff.exe                # Visual diff tool

License

This project is a fork of Valve's Source SDK 2013, licensed under the SOURCE 1 SDK LICENSE.