Overview#

The omni.hydra.rtx extension integrates NVIDIA RTX ray tracing technology into the Hydra rendering pipeline by exposing a SceneRenderer plugin as a USD/Hydra RenderDelegate. This extension serves as the bridge between Omniverse’s USD-based scene representation and RTX’s high-performance ray tracing capabilities, enabling real-time path-traced rendering within USD workflows.

The extension registers an RTX-based Hydra engine factory that creates render delegates capable of leveraging RTX hardware acceleration for lighting, shadows, reflections, and global illumination. By implementing the Hydra RenderDelegate interface, it allows RTX rendering to work seamlessly with any USD scene while maintaining compatibility with Hydra’s multi-renderer architecture.

RTX Integration#

The extension incorporates multiple RTX subsystems to provide comprehensive ray tracing functionality. The ray tracing pipeline includes dedicated plugins for core raytracing operations, denoising (including OptiX-based denoising), rasterization fallback, and post-processing effects. Material handling is enhanced through RTX-specific material databases and MDL translation capabilities that optimize material evaluation for ray tracing workloads.

Advanced features include synthetic data generation for machine learning applications, multi-GPU management for scaling across multiple RTX devices, and RTX sensor simulation for autonomous vehicle and robotics workflows. The extension also provides camera ISP simulation and ReShade integration for additional post-processing capabilities.

RTX Sensor Execution-Phase Batching#

RTX Sensor 3.0 assigns a distinct render-graph execution phase to each stage of a sensor trace by default. This allows the render graph to batch equivalent stages from multiple sensors while preserving the required stage order within each sensor:

openTrace
batchBegin
ray dispatch
batchEnd
closeTrace

Sensor models can opt out by returning shareAllocatedBuffers = true from getModelRequirements(). This keeps every stage on execution phase 0 and preserves the legacy, unphased submission behavior. Models should request this only when they depend on mutable allocated buffers that cannot safely be used with execution-phase batching.

Compatibility Setting#

The Boolean setting /rtx/rtxsensor/forceShareAllocatedBuffers defaults to false and applies to RTX Sensor 3.0. When enabled, it overrides the sensor model requirement and forces shareAllocatedBuffers = true. It is provided as a compatibility and diagnostic escape hatch for sensor models that have not yet been validated with execution-phase batching.

The setting should normally be specified at application startup:

--/rtx/rtxsensor/forceShareAllocatedBuffers=true

Leaving the setting disabled does not prevent an individual sensor model from requesting shared-buffer compatibility mode through its own requirements.