Providing Highly Detailed Geometry for Cloud and Edge Real-Time Rendering

Information

Abstract

Mesh shading was recently introduced as a topical GPU feature in the NVIDIA Turing and AMD RDNA2 GPU architectures, offering an alternative pathway for executing the transformation, generation and augmentation of geometry for hardware rasterization. Future trends in game development will rely on mesh shading and “meshlets”, using highly detailed meshes with deep level of detail hierarchies. Particularly powerful applications of meshlets include arbtirary culling and subdivision methods. Furthermore, advanced pre-computation include visibility and lighting information that can be stored on a per-meshlet basis, thus promoting the compression of attributes through quantization and the acceleration of computations via hierarchical processing.

Although meshlets can be comprised from arbitrary assemblages of primitives, their benefits are highest when meshlet formation is done in a way that already takes the usecase into account. Individual formation procedures can be defined in order to achieve specific goals. As an example, we may generate meshlets that are optimized for global illumination techniques, by minimizing their curvature and variance in material coefficients. Incoming light can then be ray-traced and cached per meshlet, along with view-dependent variance encoded in a discretized data structure. More uniform meshlets thus require less data transferred for accurately approximating their global illumination, reducing the consumption of critical memory bandwidth. We may also partition entire scenes into meshlets that foster fast visibility culling for large groups of primitives, without transforming even a single vertex. In fact, meshlet formation policies can leverage arbitrary attributes, such as the distribution of UV coordinates, ambient occlusion or mesh topology in order to optimize them according to desired runtime criteria. Cloud gaming offers a unique opportunity for leveraging this technology at a larger scale: dedicated data storages and servers can maintain multiple copies of complex triangle meshes, each partitioned by a particular meshlet formation policy. A live monitor can react to a specific bottleneck by dynamically switching meshlets to best accommodate the current GPU resource requirements. In this talk, we will present the various possibilities for real-time rendering to benefit from mesh shading by means of optimized meshlet formation procedures.

Additional Files and Images

Additional images and videos

Additional files

Weblinks

No further information available.

BibTeX

@talk{kerbl_2021_hdg,
  title =      "Providing Highly Detailed Geometry for Cloud and Edge
               Real-Time Rendering",
  author =     "Bernhard Kerbl",
  year =       "2021",
  abstract =   "Mesh shading was recently introduced as a topical GPU
               feature in the NVIDIA Turing and AMD RDNA2 GPU
               architectures, offering an alternative pathway for executing
               the transformation, generation and augmentation of geometry
               for hardware rasterization. Future trends in game
               development will rely on mesh shading and “meshlets”,
               using highly detailed meshes with deep level of detail
               hierarchies. Particularly powerful applications of meshlets
               include arbtirary culling and subdivision methods.
               Furthermore, advanced pre-computation include visibility and
               lighting information that can be stored on a per-meshlet
               basis, thus promoting the compression of attributes through
               quantization and the acceleration of computations via
               hierarchical processing.  Although meshlets can be comprised
               from arbitrary assemblages of primitives, their benefits are
               highest when meshlet formation is done in a way that already
               takes the usecase into account. Individual formation
               procedures can be defined in order to achieve specific
               goals. As an example, we may generate meshlets that are
               optimized for global illumination techniques, by minimizing
               their curvature and variance in material coefficients.
               Incoming light can then be ray-traced and cached per
               meshlet, along with view-dependent variance encoded in a
               discretized data structure. More uniform meshlets thus
               require less data transferred for accurately approximating
               their global illumination, reducing the consumption of
               critical memory bandwidth. We may also partition entire
               scenes into meshlets that foster fast visibility culling for
               large groups of primitives, without transforming even a
               single vertex. In fact, meshlet formation policies can
               leverage arbitrary attributes, such as the distribution of
               UV coordinates, ambient occlusion or mesh topology in order
               to optimize them according to desired runtime criteria.
               Cloud gaming offers a unique opportunity for leveraging this
               technology at a larger scale: dedicated data storages and
               servers can maintain multiple copies of complex triangle
               meshes, each partitioned by a particular meshlet formation
               policy. A live monitor can react to a specific bottleneck by
               dynamically switching meshlets to best accommodate the
               current GPU resource requirements. In this talk, we will
               present the various possibilities for real-time rendering to
               benefit from mesh shading by means of optimized meshlet
               formation procedures.",
  month =      jul,
  event =      "InnovWave 2021",
  location =   "online",
  keywords =   "cloud, real-time, rendering",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2021/kerbl_2021_hdg/",
}