RTX - Real-Time mode

Omniverse RTX Renderer provides the RTX - Real-Time ray tracing mode which allows rendering more geometry than traditional rasterization methods as well as physically-based materials at a high fidelity, in real-time.

In RTX - Real-Time mode, the renderer performs a series of separate passes that compute the different lighting contributions (for example: ray-traced ambient occlusion, direct lighting with ray-traced shadows, ray-traced indirect diffuse global illumination, ray-traced reflections, ray-traced translucency and subsurface scattering). Each pass is separately denoised, and the results are composited.

This mode is slightly less accurate than RTX – Interactive (Path Tracing) mode, due to using various shading approximations and optimizations to maintain a high framerate.

Eco Mode



Stop Rendering After This Many Frames Without Changes

If nothing is updated in the scene, rendering will be paused after this amount of frames, and will resume when a new update is triggered. This mode is useful to save power on laptops.




Frame Generation

DLSS Frame Generation boosts performance by using AI to generate more frames. DLSS analyzes sequential frames and motion data to create additional high quality frames. This feature requires an Ada Lovelace architecture GPU.

Super Resolution

DLSS Super Resolution boosts performance by using AI to output higher resolution frames from a lower resolution input. DLSS samples multiple lower resolution images and uses motion data and feedback from prior frames to reconstruct native quality images.
  • Mode

Select between Performance, Balanced or Quality modes.
  • Sharpness

Higher values produce sharper results.
  • Exposure mode

Choose between Force self evaluated, PostProcess AutoExposure, Fixed.
  • Auto Exposure Multiplier

Factor with which to multiply the selected exposure mode.

Direct Lighting



Direct Lighting

Enable direct light contributions from light primitives.

New Denoiser (Experimental)

Enables the new experimental denoiser.


When disabled, lights will not cast shadows.

Sampled Direct Lighting Mode

Enables Sampled Lighting which scales well with many lights (10 or more), but is less temporarly stable due to denoising.

Auto-enable Sampled Direct Lighting Above Light Count Threshold

Automatically enables Sampled Direct Lighting when the light count is greater than the Light Count Threshold.
  • Light Count Threshold

Light count threshold above which Sampled Direct Lighting is automatically enabled.

Non-Sampled Direct Lighting


Shadow Samples per Pixel

Higher values increase the quality of shadows at the cost of performance.

Lower Resolution Shadows Denoiser

Reduces the shadow denoiser resolution to gain performance at the cost of quality.

Dome Lighting

Enables dome light contribution to diffuse BSDFs if Dome Light mode is IBL.
  • Dome Lighting in Reflections

Enables dome light sampling in reflections.
  • Dome Light Samples per Pixel

Higher values increase dome light sampling quality at the cost of performance.

Sampled Direct Lighting


Samples per Pixel

Higher values increase the direct lighting quality at the cost of performance.

Max Ray Intensity

Clamps the brightness of a sample, which helps reduce fireflies, but may result in some loss of energy.

Reflections: Light Samples per Pixel

Higher values increase the reflections quality at the cost of performance.

Reflections: Max Ray Intensity

Clamps the brightness of a sample, which helps reduce fireflies, but may result in some loss of energy.

Firefly Filter

Sampled Direct Lighting Denoiser firefly filter mode. Clamps overly bright pixels to a maximum value. Choose the filter type (None, Median, RCRS).

History Clamping

Sampled Direct Lighting Denoiser history clamping. Reduces temporal lag.

Denoiser Iterations

Sampled Direct Lighting Denoiser iteration count. Number of times the frame is denoised.

Mesh-Light Sampling

Enables direct illumination sampling of geometry with emissive materials.

Indirect Diffuse Lighting



Indirect Diffuse GI

Enables Diffused Global Illumination.
  • Samples per Pixel

Higher values result in smoother looking GI.
  • Max Bounces

More bounces allow more potential GI contribution in occluded areas.
  • Intensity

Controls the intensity of the GI contribution evenly across the occluded area.
  • Denoising Mode

Denoising algorithm selection.

Ambient Occlusion

Enables ambient occlusion.
  • Ray Length

The radius around the intersection point which the ambient occlusion affects.
  • Minimum Samples per Pixel

Minimum number of samples per frame for ambient occlusion sampling.
  • Maximum Samples per Pixel

Maximum number of samples per frame for ambient occlusion sampling.
  • Denoiser

Allows for increased AO denoising at the cost of more blurring.

Ambient Light Color

Color of the global environment lighting.

Ambient Light Intensity

Brightness of the global environment lighting.

Denoising Mode: NVRTD


Kernel Radius

Controls the spread of local denoising area. Higher values results in smoother GI.

Iteration Count

The number of denoising passes. Higher values results in smoother looking GI.

Max History Length

Controls latency in GI updates. Higher values results in smoother looking GI.





Enable reflections.

Roughness Cache Threshold

Roughness threshold for approximated reflections. Higher values result in better quality, at the cost of performance.

Max Reflection Bounces

Number of bounces for reflection rays.





Enables translucency for specular transmissive surfaces such as glass.

Max Refraction Bounces

Number of bounces for refraction rays.

Reflection Seen Through Refraction

When enabled, reflection seen through refraction is rendered. When disabled, reflection is limited to first bounce only. More accurate, but worse performance.

Secondary Bounce Reflection Throughput Threshold

Threshold below which reflection paths due to fresnel are no longer traced. Lower values result in higher quality at the cost of performance.

Fractional Cutout Opacity

Enables fractional cutout opacity values resulting in a translucency-like effect similar to alpha-blending.

Depth Correction for DOF

Improves DoF for translucent (refractive) objects, but can result in worse performance.

Motion Vector Correction (experimental)

Enables motion vectors for translucent (refractive) objects, which can improve temporal rendering such as denoising, but can result in worse performance.

World Epsilon Threshold

Treshold below which image-based reprojection is used to compute refractions. Lower values result in higher quality at the cost performance.

Roughness Sampling (experimental)

Enables sampling roughness, such as for simulating frosted glass, but can result in worse performance.

Subsurface Scattering



Max Sample Per Frame

Max samples per frame.


Enables denoising for the subsurface scattering.


Enables transmission of light through the medium, but requires additional samples and denoising.
  • BSDF Sample Count

Transmission sample count per frame.
  • Samples Per BSDF Sample

Transmission samples count per BSDF Sample. Samples per pixel per frame = BSDF Sample Count * Samples Per BSDF Sample.
  • Screen-Space Fallback Threshold

Transmission threshold for screen-space fallback.
  • Half-Resolution Rendering

Enables rendering transmission in half-resolution to improve performance at the expense of quality.
  • Sample Guiding

Enables transmission sample guiding, which may help with complex lighting scenarios.
  • Denoiser

Enables transmission denoising.





Enables simulation of caustics generated for area lights whic have the ‘Enable Caustics’ enabled.

Photon Count Multiplier

Factor multiplied by 1024 to compute the total number of photons to generate from each light.

Photon Max Bounces

Maximum number of bounces to compute for each light/photon path.

Filter Iterations

Number of iterations for the denoiser applied to the results of the caustics tracing pass.

Global Volumetric Effects



Accumulation Frames

Number of frames samples accumulate over temporally. High values reduce noise, but increase lighting update times.

Depth Slices Count

Number of layers in the voxel grid to be allocated. High values result in higher precision at the cost of memory and performance.

Pixel Density

Higher values result in higher fidelity volumetrics at the cost of performance and memory (depending on the # of depth slices).

Slice Distribution Exponent

Controls the number (and relative thickness) of the depth slices.

Inscatter Blur Sigma

Sigma parameter for the Gaussian filter used to spatially blur the voxel grid. 1 = no blur, higher values blur further.

Inscatter Dithering Scale

The scale of the noise dithering. Used to reduce banding from quantization on smooth gradients.

Spatial Sample Jittering Scale

Scales how far light samples within a voxel are spatially jittered: 0 = only from the center, 1 = the entire voxel’s volume.

Temporal Reprojection Jittering Scale

Scales how far to offset temporally-reprojected samples within a voxel: 0 = only from the center, 1 = the entire voxel’s volume. Acts like a temporal blur and helps reduce noise under motion.

Use 32-bit Precision

Allocate the voxel grid with 32-bit per channel color instead of 16-bit. This doubles memory usage and reduces performance, generally avoided.

Flow Sampling

Samples density from Flow grid.
  • Min Layer

Min Flow layer to sample, inclusive.
  • Max Layer

Max Flow layer to sample, inclusive.
  • Density Scale

Scale to apply to sampled Flow density.
  • Density Offset

Offset applied after scale. 1.0 allows existing fog to continue working normally.


Multi-GPU rendering in RTX - Real-Time mode splits the image to render as tiles which are dispatched to the GPUs and then composited into a single image.

The primary GPU performs various tasks, such as: rendering pixels, sample aggregation, denoising, post processing, UI rendering. The default GPU 0 Weight value is usually ideal.




Enables using multiple GPUs (when available). This splits the rendering of the image into a large tile per GPU with a small overlap region between them.

Automatic Tiling

Automatically determines the image-space grid used to distribute rendering to available GPUs. The image rendering is split into a large tile per GPU with a small overlap region between them. Note that by default not necessarily all GPUs are used. The approximate number of tiles is viewport resolution divided by the Minimum Megapixels Per Tile setting, since at low resolution small tiles distributed across too many devices decreases performance due to multi-GPU overheads. Disable automatic tiling to manually specify the number of tiles to be distributed across devices.
  • Minimum Megapixels Per Tile

The minimum number of Megapixels each tile should have after screen-space subdivision.

Tile Count

Number of tiles to split the image into. Usually this should match the number of GPUs, but can be less.

Tile Overlap (Pixels)

Width, in pixels, of the overlap region between any two neighboring tiles.

GPU 0 Weight

This normalized weight can be used to decrease the rendering workload on the primary device for each viewport in relation to the other secondary devices, which can be helpful for load balancing in situations where the primary device also needs to perform additional expensive operations such as denoising and post-processing.


Execute per-device render command recording in separate threads.


For efficiency’s sake, in some contexts rendering will switch to single-GPU automatically until conditions change to warrant multi-GPU rendering, for example when rendering at low resolution.

Multi-GPU rendering is enabled by default if the system has multiple NVIDIA RTX-enabled GPUs of the same model. GPUs which don’t support ray tracing are skipped automatically.

If the GPU models are not identical, multi-GPU can be enabled with the command line: --/renderer/multiGpu/enabled=true​

Per-GPU memory usage is limited to 48GB. The GPU with the lowest memory capacity will limit the amount of memory the other GPUs can leverage. You can disable the lower-capacity GPU to avoid this limitation.

To limit the maximum number of GPUs, Omniverse Kit must be launched with the following argument: --/renderer/multiGpu/maxGpuCount


SLI mode is unstable and should be globally disabled in the NVIDIA control panel for multi-GPU. This will be addressed in future releases.