RTX Mega Geometry

· Change Log · Quick Start ·

Zorah courtyard (RTX 5090 · 4K @ DLSS Quality)
15.5 ms/frame | 56M unique, 778M instanced triangles | 1.5 GB Resident Mesh | 2.3 GB CLAS

Overview

RTX Mega Geometry (RTX MG) is a DX12 and Vulkan code sample that shows how to build ray-tracing acceleration structures for high density geometry via dynamic tessellation of subdivision surfaces and streaming of pre-baked triangle clusters. It contains a reference HLSL path tracing demo app that can be used as a learning tool to begin integration with your own graphics engine.

The sample demonstrates two geometry paths in one renderer:

• Cluster LOD — pre-baked triangle clusters are selected per frame from a continuous LOD hierarchy and streamed into VRAM on demand.
• Cluster tessellation — Catmull-Clark subdivision surfaces are tessellated and displaced on the fly, and their acceleration structure is rebuilt every frame from cluster templates.

The two paths share one material table, one TLAS and one scene graph, and a single scene may use both.

Features

• Real-time adaptive sampling of Catmull-Clark limit surfaces
• Tessellation of subdivision surfaces into clusters and displacement mapping
• Continuous cluster LOD with on-demand geometry streaming, for scenes whose source geometry does not fit in VRAM
• BLAS sharing, caching and merging to keep acceleration-structure build cost flat as instance counts rise
• Cluster BVH build using NVAPI for DX12, and the VK_NV_cluster_acceleration_structure extension for Vulkan.
• Hierarchical Z-buffer for reducing detail in occluded geometry
• DLSS-RR (Ray Reconstruction) denoising and upscaling

Running the Sample

Run rtxmg_demo.exe from the bin folder. Model paths are relative to assets/ unless given in full.

Zorah (Cluster LOD)

Zorah Textured glTF is an export of the NVIDIA RTX Kit sample with 1.6 billion unique triangles (18.9 billion instanced), 2034 meshes, 4357 textures. It renders on the Cluster LOD path.

zorah_textured_public.v1.7z (70 GB download; 31 GB mesh data + 48 GB textures when extracted)

rtxmg_demo.exe -mf <path>/zorah_textured_public.v1.scene.json

Note

The first load bakes the LOD hierarchy for all 2,852 geometries in parallel and can require up to 64 GB of RAM at peak. The bake automatically caps its worker thread count based on available memory; pass --bake-workers <n> to set it explicitly. Each geometry shard is checkpointed on completion, so the bake can be interrupted and will resume from where it left off on the next launch.

Amy Kitchen Set (Cluster Tessellation)

A subdivision-surface scene with displacement mapping, tessellated every frame on the Cluster Tessellation path.

rtxmg_demo.exe -mf subdivision/amy_kitchenset.scene.json

Amy A Beautiful Game (Combined)

Both geometry paths in one scene, sharing a single material table, scene graph and TLAS.

rtxmg_demo.exe -mf subdivision/amy_abeautifulgame.scene.json

Once a scene is up, the Quick Start guide covers camera controls, the UI panels, the Inspector, the Profiler, and the VRAM Budget window; see User Interface below for the short version.

Requirements

To Run:

• Windows 10
• NVIDIA RTX GPU (10 GB VRAM or greater)
• GeForce Game Ready Driver 570 or later
• DirectX Raytracing 1.1 API or later

The Cluster LOD path sizes its pools from fixed budgets: 2048 MB of streamed geometry and 2048 MB of CLAS by default, plus a 4096 MB material-texture budget (--texture-budget-mb). All three are adjustable in the VRAM Budget window — lowering any of them trades detail for VRAM rather than failing to load. See ClusterLOD.md — Streaming budgets.

To Build:

• CMake v3.31+
• Windows 10 SDK 10.0.20348.0 or later
• MSVC Compiler 19.43.34810 (Visual Studio 2022 17.13) or later
• For Windows: DirectX 12 AgilitySDK will be fetched automatically
• For Vulkan: Vulkan SDK 1.4.313 or later

Folder Structure

/bin	default CMake folder for binaries and compiled shaders
/build	default CMake folder for build files
/extern	external submoduled libraries and SDKs, including osd_lite and donut
/assets	models, textures, scene files (git submodule)
/rtxmg	RTX Mega Geometry core
/rtxmg/subdivision	Catmull-Clark topology, subdivision plans, limit-surface evaluation
/rtxmg/cluster_tess	the cluster tessellation path — tiling, fill, CLAS templates
/rtxmg/cluster_lod	the Cluster LOD path — baking, streaming, LOD traversal, BLAS reuse
/rtxmg/scene	scene graph, materials, texture loading and budgeting
/rtxmg/hiz	Hierarchical Z-buffer, shared by both paths
/demo	the rtxmg_demo reference application and its shaders
/docs	quick start, algorithm guides, debugging guide, images

Build

At the moment, only Windows builds are supported. Both the D3D12 and the Vulkan back-end are supported there; pass -vk to run on Vulkan.

1. Clone the repository with all submodules recursively:

   git clone --recurse-submodules https://github.com/NVIDIA-RTX/rtxmg.git

2. Use CMake to configure the build and generate the project files.

   cd rtxmg
   cmake CMakeLists.txt -B ./build
   

3. Build the solution generated by CMake in the ./build/ folder.

   For example, if using Visual Studio, open the generated solution build/rtxmg.sln and build it.

4. Select and run the rtxmg_demo project. Binaries get built to the bin folder. Media assets are loaded from the assets folder.

User Interface

Once the application is running, most of the SDK features can be accessed via the UI window on the left hand side. The UI is self-documenting : hover the mouse over widgets to read tool-tips. See the Quick Start guide for more details.

Camera can be moved using W/S/A/D keys and rotated by dragging with the left mouse cursor. Holding the Alt key and left-click dragging will orbit the camera around the scene's center.

Contact

RTX Mega Geometry is under active development. Please report any issues directly through GitHub issue tracker, and for any information, suggestions or general requests please feel free to contact us at rtxmg-sdk-support@nvidia.com

Related RTX Mega Geometry Repositories

Vulkan

• vk_animated_clusters : VK_NV_cluster_acceleration_structure for animated content and general introduction to the new extension
• vk_tessellated_clusters : adaptive triangle tessellation and displacement using VK_NV_cluster_acceleration_structure
• vk_lod_clusters : The source for this SDK's implementation of Cluster LOD — LOD system and streaming using VK_NV_cluster_acceleration_structure.
• vk_partitioned_tlas update the TLAS of large dynamic scenes with VK_NV_partitioned_acceleration_structure

Citation

If you use RTX Mega Geometry in a research project leading to a publication, please cite the project.

BibTex:

@online{RTX MG,
   title   = {{{NVIDIA}}\textregistered{} {RTX Mega Geometry}},
   author  = {{NVIDIA}},
   year    = 2025,
   url     = {https://github.com/NVIDIA-RTX/rtxmg.git},
   urldate = {2025-02-06},
}

License

See LICENSE.txt

This project includes third-party software, and fetches and downloads more of it at configure time. Each component remains under its own licence; they are listed in notice.txt.