TECH
NVIDIA's rendering technologies provide the foundation for Founder Engine. I chose them to work on demanding materials and environments while keeping access to the renderer underneath. This section explains what each integration contributes, why I use it and where its limitations have affected development.
The NVIDIA SDKs supply the rendering foundations. My work connects and extends them through the engine's materials, scene systems and authoring tools. Different rendering modes use different combinations of features.
RTX Path Tracing
RTXPT provides the path-tracing foundation: light transport, material evaluation and the data needed for real-time reconstruction. Donut and NVRHI supply the application and graphics abstraction layers underneath.
Why I chose it
I wanted a renderer I could inspect and extend while building an editor around it. RTXPT gave me an established starting point for investigating skin, hair and procedural worlds without first having to recreate every part of a path tracer.
In Founder Engine
Founder Engine adds its own scene workflow, editor, terrain, vegetation, water and animation systems. Path-traced, hybrid RTX and raster PBR are separate modes, with different costs and feature combinations.
NVIDIA documentationDLSS Super Resolution & Ray Reconstruction
Super Resolution reconstructs a higher-resolution image from a lower internal resolution. Ray Reconstruction resolves sparsely sampled ray-traced lighting using the renderer's material, depth and motion information.
Why I chose it
Interactive feedback matters when adjusting skin, hair and lighting. DLSS lets me judge those changes while moving the camera instead of waiting for a clean accumulated image. It also makes the quality of the renderer's motion vectors and surface guides immediately relevant.
In Founder Engine
Ray Reconstruction is my main real-time character-lookdev path. Fine hair coverage, smearing, ghosting and history resets remain active work; reconstruction does not remove the need to supply correct geometry and motion data.
NVIDIA documentationStreamline, Frame Generation & Reflex
Streamline connects supported NVIDIA features through shared frame data and resource tagging. Founder Engine exposes DLSS Frame Generation and Reflex through that integration.
Why I chose it
I chose a common integration boundary for reconstruction, presentation and latency controls. Frame Generation can smooth presentation on supported hardware, while Reflex helps manage latency alongside it.
In Founder Engine
Generated frames do not accelerate the simulation or the underlying rendered frame. These are optional, hardware-dependent paths, so a frame counter needs the active configuration to be meaningful.
NVIDIA documentationRTX Direct Illumination / ReSTIR
RTXDI reuses useful light samples across space and time. The engine includes selectable ReSTIR direct- and indirect-lighting paths, with ReGIR controls for light sampling.
Why I chose it
A scene can contain far more potential light sources than can be evaluated per pixel. I wanted to spend a limited ray budget on useful samples rather than treating all candidates equally.
In Founder Engine
The RTXDI integration is selectable. ReSTIR is disabled in the current main DLSS Ray Reconstruction configuration; the listed technologies are not all enabled at once.
NVIDIA documentationNVIDIA Real-Time Denoisers
NRD supplies REBLUR and RELAX paths for denoising ray-traced signals. Surface and motion information guide the accumulation of samples while preserving lighting detail.
Why I chose it
Alternative denoisers help me distinguish a shading problem from a history or reconstruction problem. Comparing their response is useful when detail disappears or a moving surface leaves a trail.
In Founder Engine
NRD remains available alongside the primary DLSS Ray Reconstruction path. The chosen reconstruction mode determines which denoiser runs.
NVIDIA documentationRTX Character Rendering
The RTXCR Material Library provides skin-scattering functions and strand-hair shading used in the character work. Subsurface transport allows light to spread beneath the skin, while hair requires a response designed for fibres.
Why I chose it
I wanted skin to retain its complexion while responding to light through scattering, and hair to behave as fibres rather than an ordinary glossy surface. RTXCR provided a foundation I could connect to imported character data and editable material controls.
In Founder Engine
The skin work separates diffuse normal detail from the reflection frame. Generated strand grooms use the RTXCR far-field hair model; imported hair cards retain their own path. These are distinct material and geometry choices.
NVIDIA documentationRTX Mega Geometry
Clustered acceleration structures support detailed ray-traced geometry. Founder Engine applies this approach to custom terrain, mesh and foliage systems, including terrain micro-displacement.
Why I chose it
I wanted surface relief to affect silhouettes and ray intersections, not only the shading normal. Clusters offer a way to organise that geometry and decide what detail to build or keep resident as the view changes.
In Founder Engine
This is custom engine integration around NVIDIA's cluster-geometry capabilities, not an unchanged RTXMG sample. Geometry budgets, residency, streaming and hardware support remain central constraints.
NVIDIA documentationOpacity Micromaps & Shader Execution Reordering
Opacity Micromaps encode opacity-test information for ray traversal, especially useful for leaf and needle cards. Shader Execution Reordering groups compatible shader work to improve execution coherence.
Why I chose it
Forests combine many cutout surfaces with rays that encounter different materials. I chose these techniques to investigate that cost while preserving the authored foliage silhouettes.
In Founder Engine
Both paths depend on API, driver and hardware support. Micromap baking also has a preparation cost; bake-queue stalls have been part of development, so enabling the feature is not automatically a net improvement.
NVIDIA documentationRTX Neural Texture Compression
RTXNTC compresses related material textures into neural texture sets. Founder Engine uses inference on load: the bundle is decompressed during material loading, and existing shaders receive ordinary GPU textures.
Why I chose it
I wanted to explore smaller texture assets without replacing the existing PBR sampling workflow. Load-time inference makes that experiment possible while keeping material evaluation familiar and inspectable.
In Founder Engine
Smaller files do not imply the same reduction in runtime VRAM. This path does not run neural inference at each texture lookup. Early roughness mismatches also showed that preserving packed channel meanings matters as much as compression quality.
NVIDIA documentationRTX Texture Filtering
RTX Texture Filtering provides stochastic filtering routines in the path tracer. Sampling is distributed over time and resolved through reconstruction rather than judged from one frame alone.
Why I chose it
I wanted filtering that works with temporal reconstruction and detailed material textures. It is another part of making the texture response stable as the camera moves, alongside correct motion and reconstruction inputs.
In Founder Engine
Filtering and compression solve different problems. RTXTF does not turn the RTXNTC load-time path into inference on sample; the final appearance still depends on reconstruction, motion and the source material.
NVIDIA documentation