Traverse Research / Rendering & GPU R&D

What comes after the next frame?

Learned materials. Neural representations. New GPU architectures. We work with hardware and engine partners to turn promising research into implementations you can investigate, and we benchmark the results: Evolve, our GPU and AI benchmark, is built by this team.

Explore the research Talk to an engineer

Representation → implementation → measurement

Traverse × Samsung / Mobile ray tracing

A public look at our work with Samsung.

Samsung shared this Xclipse GPU demonstration on its official channel. Built in our Breda rendering framework, it shows our mobile ray-tracing collaboration in action.

Explore the Samsung collaboration

Samsung / Xclipse GPU: Hyper-realistic Ray TracingWatch on YouTube

Explore the layers of the problem.

From the material representation to the hardware running it. Follow the question across the stack.

NeuBTF material representations from Traverse’s published research.

Neural rendering

Radiance caching, neural materials, texture compression and GPU inference. Built into the renderer, from training to the frame.

Explore Neural rendering

Neural radiance cache at the primary vertex, from our Evolve capture. Internal debug build.

Neural radiance caching

Neural lighting research published with Samsung at SIGGRAPH Asia 2025. Training, inference and path tracing designed together.

Explore Neural radiance caching

A workload render from Evolve, our GPU and AI benchmark.

Benchmarking and GPU workload research

Repeatable workloads that reveal architectural trade-offs, and Evolve, the benchmark we build and sell.

Explore Benchmarking and GPU workload research

Representative Breda material/rendering capture. No compression ratio is claimed for this image.

Neural texture compression

Learned texture representations, evaluated from training through reconstruction and GPU integration.

Explore Neural texture compression

Per-material validation loss from a neural-material training run in the Breda framework, a panel to a material.

GPU inference and training

Neural workloads that belong inside the frame: operators, model interchange, resource flow and GPU execution.

Explore GPU inference and training

Traverse volumetric path tracer, with multiple scattering and up to 16 light bounces.

Rendering engineering partner

Implementations that connect light transport, materials and hardware behaviour.

Explore Rendering engineering partner

NeuBTF material research, published by Traverse Research.

Neural materials

Represent material appearance with learned functions, and make those functions practical to render and stream.

Explore Neural materials

Three ray-tracing micro-benchmarks from our pre-silicon suite, each presenting the acceleration structure with a different pattern of overlap and geometry distribution. Magenta is a ray that hits nothing.

GPU pre-silicon validation

Ray-tracing micro-benchmarks that expose traversal and acceleration-structure behaviour on hardware that does not exist yet.

Explore GPU pre-silicon validation

Breda AS Playground: builder settings, live BVH statistics, a GPU frame breakdown and an intersection heatmap of where traversal work lands.

Acceleration structures

BVH construction, refit and traversal quality, compared strategy by strategy in our own playground and taken through to a driver integration.

Explore Acceleration structures

Strand rendering on a mobile GPU, the same asset rendered twice and set side by side - which is how a change is shown not to have cost image quality.

Hair and strand rendering

Fine geometry, capsule clustering and workload organisation for real-time strand rendering.

Explore Hair and strand rendering

Ray-traced Gaussians

Gaussian scene representations evaluated with ray tracing, lighting, shadows and image-quality comparisons.

Explore Ray-traced Gaussians

The connection between sensor data, rendering passes and the simulation system.

Rendering for simulation

Real-time graphics engineering for simulation systems, connecting sensor data flow, image quality and GPU cost.

Explore Rendering for simulation

NVIDIA’s Zorah scene, rendered in Traverse’s Breda framework.

Micro-polygon rendering

Detailed geometry, visibility buffers, mesh shading and asset preparation investigated as one rendering pipeline.

Explore Micro-polygon rendering

Volumetric rendering

Animated volumes, heterogeneous media and clouds, with Rust tools for the data that feeds the renderer.

Explore Volumetric rendering

Neural reconstruction schematic: inputs, inference and image output.

Neural reconstruction and upscaling

For silicon and driver teams building a reconstruction technology: independent implementation in a real renderer, evaluation against the competing upscalers, and the SDK and bindings that get developers adopting it.

Explore Neural reconstruction and upscaling

Indirect light through an interior, rendered in the Breda framework.

Light transport and global illumination

Path tracing, ReSTIR, dynamic diffuse GI and neural radiance caching connected through a working rendering framework.

Explore Light transport and global illumination

Evolve, our GPU and AI benchmark, is where much of this practice was learned.

SDK development for hardware vendors

The library a developer actually meets: C++ or Rust SDKs, bindings and small standalone samples that show what your hardware does well, built by a team that has been on the integrating end of a great many of them.

Explore SDK development for hardware vendors

Look inside the research

From a research question to something you can inspect.

NeuBTF investigates learned material appearance. Explore the published project, then see how rendering, neural materials and GPU engineering connect in our wider work.

Explore NeuBTF

SIGGRAPH Asia 2022 / Posters

NeuBTFNeural material appearanceRead the published research

Learned material appearance, with published quality and compression comparisons.

Traverse Research + Traverse Games

Let’s turn a good question into working software.

Start a conversation