{"id":11828,"date":"2018-09-17T06:00:19","date_gmt":"2018-09-17T13:00:19","guid":{"rendered":"https:\/\/developer.nvidia.com\/blog\/?p=11828"},"modified":"2023-08-02T08:21:12","modified_gmt":"2023-08-02T15:21:12","slug":"introduction-turing-mesh-shaders","status":"publish","type":"post","link":"https:\/\/developer.nvidia.com\/blog\/introduction-turing-mesh-shaders\/","title":{"rendered":"Introduction to Turing Mesh Shaders"},"content":{"rendered":"

The Turing architecture introduces a new programmable geometric shading pipeline through the use of mesh shaders<\/strong>. The new shaders bring the compute programming model to the graphics pipeline as threads are used cooperatively to generate compact meshes (meshlets<\/strong>) directly on the chip for consumption by the rasterizer. Applications and games dealing with high-geometric complexity benefit from the flexibility of the two-stage approach, which allows efficient culling, level-of-detail techniques as well as procedural generation.<\/p>\n

This blog introduces the new pipeline and gives some concrete examples in GLSL for OpenGL or Vulkan rendering. The new capabilities are accessible through extensions in OpenGL and Vulkan, and using DirectX 12 Ultimate<\/a>.<\/p>\n

Most of the following content is taken from this recorded presentation<\/a>, for which the full slide deck will be available at a later date.<\/p>\n

\n

1\u00a0 <\/span>Mesh Shading Pipeline<\/a>
\n2\u00a0 <\/span>Meshlets and Mesh Shading<\/a>
\n3\u00a0 <\/span>Pre-Computed Meshlets<\/a>
\n3.1\u00a0 <\/span>Data Structures<\/a>
\n3.2\u00a0 <\/span>Rendering Resources and Data Flow<\/a>
\n3.3\u00a0 <\/span>Cluster Culling with Task Shader<\/a>
\n4\u00a0 <\/span>Conclusion<\/a>
\n5\u00a0 <\/span>References<\/a><\/p>\n<\/div>\n

Motivation<\/h1>\n

The real world is a visually rich, geometrically complex place. Outdoor scenes in particular can be composed of hundreds of thousands of elements (rocks, trees, small plants, etc.). CAD models present similar challenges with both complex shaped surfaces as well as machinery made of many small parts. In visual effects large structures, for example spaceships, are often detailed with “greebles”. Figure 1 shows several examples where today\u2019s graphics pipeline with vertex, tessellation, and geometry shaders, instancing and multi draw indirect, while very effective, can still be limited when the full resolution geometry reaches hundreds of millions of triangles and hundreds of thousands of objects.<\/p>\n

\"Turing<\/a>
Figure 1. The need for increasing realism drives massive increases in geometric complexity.<\/figcaption><\/figure>\n

Other use-cases not shown above include geometries found in scientific computing (particles, glyphs, proxy objects, point clouds) or procedural shapes (electric engineering layouts, vfx particles, ribbons and trails, path rendering).<\/p>\n

In this post we look at mesh shaders<\/em> to accelerate rendering of heavy triangle meshes. The original mesh is segmented into smaller meshlets<\/strong> as figure 2 shows. Each meshlet ideally optimizes the vertex re-use within it. Using the new hardware stages and this segmentation scheme, we can render more geometry in parallel while fetching less overall data.<\/p>\n\n\n\n
Figure 2. : Large meshes can be decomposed into meshlets, which are rendered by mesh shaders.<\/caption>\n
\"\"<\/a><\/td>\n\"\"<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

For example CAD data can reach tens to hundreds of millions of triangles. Even after occlusion culling<\/a>\u00a0a significant amount of triangles can exist. Some fixed-function steps in the pipeline may do wasteful work and memory loads in this scenario:<\/p>\n