geometryfx screenshot

The GeometryFX library provides convenient access to compute-based triangle filtering (CTF), which improves triangle throughput by filtering out triangles that do not contribute to the final image using a compute based pre-process.

Note: GeometryFX was designed for GCN-architecture GPUs. On RDNA, triangle throughput is much less of a problem than it used to be and per-triangle culling should no longer be handled by the application. Cluster culling remains unaffected though, but given the minimal amount of code required for it there’s no need to pull in the GeometryFX library for that.

How It works

GeometryFX improves the rasterizer efficiency by culling triangles that do not contribute to the output in a pre-pass. This allows the full chip to be used to process geometry, and ensures that the rasterizer only processes triangles that are visible.

A good use case for the GeometryFX library is depth-only rendering of opaque geometry – for example, in shadow maps:

  • Depth-only rendering leaves most compute units idle, which can be used by GeometryFX.
  • Opaque geometry has no ordering requirements, so GeometryFX can cull triangles in arbitrary order and regroup/split draw calls.
  • All geometry can be rendered using the same vertex shader, which allows the GeometryFX library to merge draw calls for maximum efficiency.

At its core, GeometryFX works by generating an intermediate index buffer which consists of visible triangles only. Intermediate buffers are reused as much as possible to minimize memory usage. GeometryFX also buffers up draw calls to execute the filtering on one batch while the previous batch is being rendered, allowing the filtering to overlap with the actual draw call.

The library makes heavy use of multi-draw-indirect. This is a D3D11 driver extension exposed through the AMD GPU Services (AGS) library. It allows multiple draw calls to be prepared on the GPU and executed with a single API call. For more information on AGS, visit GPUOpen’s AGS page.


  • AMD Radeon™ GCN-based GPU (HD 7000 series or newer)
    • Or other DirectX® 11 compatible discrete GPU with Shader Model 5 support
  • 64-bit Windows® 7 (SP1 with the Platform Update), Windows 8.1, or Windows 10
  • Visual Studio 2012, Visual Studio 2013, or Visual Studio 2015

Technical Blogs

Related GPUOpen blog posts