This technical deep dive was written by Cristiano Siqueira, Manu Mathew Thomas, Miroslaw Pawlowski, and Anton Sochenov as part of their research efforts at the Visual Compute and Graphics lab, within Intel Labs
Highlights:
- This blog details the specific technical considerations required to address the challenges of real-time path tracing from the scene's perspective.
- Jungle Ruins consolidates multiple technical challenges – including high-frequency detail, massive instancing, and alpha testing – into a single, unified workload.
- Real-time path tracing offers many benefits to the lighting and rendering pipeline, such as accurate dynamic lighting, simplified asset authoring, and consistency between authoring and final rendering.
- The insights gained from building the Jungle Ruins scene and our research journey offer a strong foundation for future content creation and research direction and contribute directly to the ongoing advancement of path-tracing applications in production environments.
At a high level, Jungle Ruins is a research-driven scene that explores several open challenges in real-time path tracing. These complex rendering problems are creatively combined in a visually rich experience inspired by the natural beauty and complexity of the Amazon rainforest.
We have previously explored the creative direction of Jungle Ruins in a separate blog post, found here. This blog will focus on the specific technical considerations required to address the challenges of real-time path tracing from the scene's perspective.
Figure 1. Jungle Ruins in our research framework.
Unified Workload
Jungle Ruins consolidates multiple technical challenges into a single, unified workload. This approach allows us to expose and analyze complex interactions between challenges that might not be apparent when addressed individually. Among the core challenges we unified in our workload are:
- High-frequency Detail – Rendering high-frequency details, such as fine textures or sharp edges, poses a significant challenge in complex scenes with very dense geometry, where one sample per pixel leads to aliasing and exponentially increases the denoising complexity.
Figure 2. Jungle Ruins’ high-frequency detail captured in our research framework.
Massive Instancing – Massive instancing is essential for representing realistic forests and vegetation density, often involving millions of instances. This scale introduces significant challenges in memory and acceleration structure management in real-time path tracing. Each unique geometry requires separate Bottom-Level Acceleration Structures (BLAS), leading to increased memory consumption and build times. Furthermore, the Top-Level Acceleration Structures (TLAS) become substantially large, impacting traversal efficiency and overall performance.
Figure 3. Jungle Ruins’ dense vegetation instancing captured in Blender.
Alpha Testing – In path tracing, alpha testing for vegetation is computationally demanding, as it requires evaluating opacity maps at every ray intersection, significantly increasing memory access for high-resolution textures and ray traversal time through multiple semi-transparent layers. This put considerable strain on performance in scenes with dense, intricate foliage. An alternative approach is to use fully modeled leaf geometry without opacity maps, eliminating the computational overhead of alpha testing. However, this approach may dramatically increase the unique and total triangle counts, depending on the shape and complexity of the leaves, exceeding technical budgets and GPU memory constraints, thus requiring careful consideration.
Figure 4. Wireframe comparison of full geometry leaves and opacity map-driven leaves.
Scene Management and Modularity
Scalability and modularity were central to the technical design of the Jungle Ruins scene, supporting our research goals. As the creative blog post describes, the scene covers an 8km x 8km area (~64 km2), partitioned into multiple independent tiles. This tiling method, common in game engines, was incorporated into the scene design to ensure compatibility with our research framework.
This scalable structure allowed the artist to focus on asset creation while researchers could efficiently load and experiment with individual tiles. It enabled rapid prototyping and evaluation of rendering techniques without slowing content creation. As a result, artistic development and rendering research could proceed simultaneously, seamlessly integrating into the final assembled scene without requiring major adjustments.
The modular setup was critical, simplifying targeted adjustments and iterative testing. For example, individual vegetation assets or entire groups could be quickly swapped or optimized. This flexibility significantly streamlined complexity management and large-scale performance optimizations.
Figure 5. Scene/world partition visualized in random colors.
Technical Challenges and Implications
Optimizing the Jungle Ruins scene for this project presented several challenges. First, our target platform uses low-end GPUs, imposing strict technical budget requirements to meet performance goals. Additionally, our research framework supports only basic scene features, further restricting our options.
On the platform side, we prioritized asset optimization to maximize performance without compromising visuals. This involved simplifying the base geometry, using a limited number of materials, and strategically distributing texture resolutions to manage GPU memory constraints effectively. With a limited number of unique triangles, we had to choose where to spend them best. After several iteration cycles, we’ve decided to concentrate as much of that detail on the surface vegetation itself as possible, as once instanced, that is what would mostly be visible from the camera.
To further boost performance, we’ve explored rendering techniques that would bring us closer to meeting our goals. One effective approach was disabling alpha testing where it has minimal visual impact. By skipping alpha tests for geometry beyond a certain distance from the ray origin, we reduced the number of unnecessary visibility checks through complex alpha-masked surfaces. For secondary rays, a threshold of 1 meter was effective, while primary rays tolerated distances between 50 and 100 meters without noticeable artifacts. Only the pyramid and tall grass showed subtle differences. Depending on the camera position, this simple heuristic can improve total rendering time by up to 10%. That said, this technique introduced a challenge with the pyramid’s moss, where the original quad geometry contained excessive white space in its texture maps, causing a color shift issue when alpha testing was triggered as the camera moved closer or farther. To address this, we recreated the moss base geometry closely following the outlines of its opacity texture, eliminating the need for an opacity map entirely and resolving the color shift issue.
However, our framework’s limitations posed additional challenges. Its basic support for simple Physically Based Rendering (PBR) materials and direct texture inputs prevented us from leveraging tiled textures for the terrain, resulting in less detail in close-up views. The lack of a dedicated terrain rendering system in our framework meant that certain areas, like the cliff highlighted below in Figure 6, ended up with lower detail than initially planned.
Addressing these challenges required creative solutions to achieve our visual targets. By shifting our triangle budget toward high-quality vegetation, we’ve significantly enhanced the scene’s appeal, especially when paired with the realism of the real-time path tracer. Carefully scattered foliage, dense ground cover detail, and canopy trees effectively concealed terrain imperfections. Ultimately, these strategies enabled us to meet the project’s technical requirements while delivering a visually stunning result we’re proud of.
Benefits of Real-Time Path Tracing
Real-time path tracing brings many advantages to the lighting and rendering pipeline, fundamentally simplifying workflows for creators. By leveraging real-time physics-based lighting algorithms, teams can achieve high-fidelity visuals with greatly reduced manual tuning and complex workarounds from traditional real-time rendering pipelines. Key benefits include:
- Accurate Dynamic Lighting – One of the most transformative changes is eliminating pre-baked lighting. With real-time path tracing, global illumination, shadows, and reflections are computed on the fly in each frame, removing the need to generate lightmaps or other static lighting data. Artists can move lights or adjust emissive materials and immediately see accurate lighting updates instead of waiting hours for a bake. This ability to iterate lighting setups enables greater creative freedom, resulting in a much faster lighting workflow where what you see is what you get.
Figure 7. Sun timelapse with dynamic lighting.
- Simplified Asset Authoring – Real-time path tracing allows assets to be authored more straightforwardly, using physically accurate properties rather than engine-specific hacks. Since the path tracer simulates light behavior under a unified physics model, materials behave as they would in reality – whether a perfectly reflective mirror, a piece of rough metal, or translucent glass – and are all handled in the same lighting system. Artists no longer need to create special case shader tricks to account for missing lighting effects, as the algorithm inherently captures these effects. For example, baked ambient occlusion textures or overly brightening materials to compensate for lost bounced lighting become redundant. This means fewer workarounds in material setups and more time spent on artistic refinement. Path tracing’s generality ensures assets can be built with true PBR values and look correct. The asset pipeline is cleaner and more physically grounded, benefiting artists and shader developers.
- Consistency Between Authoring and Final Render – Another benefit of real-time path tracing is greatly narrowing the gap between how content is previewed during creation and how it appears in the final render. In a typical pipeline, an artist might see their asset under one lighting setup in a 3D DCC or offline renderer, only to have it look different in the game engine due to differences in rendering algorithms. This often leads to multiple iteration loops between artists and engineers trying to reconcile the differences. With a path-traced engine, lighting and shading adhere to physical correctness at runtime, which brings us much closer to the ground truth. Many studios use path tracers as a reference to verify their real-time lighting, and with hardware advancements, these models are becoming fast enough for interactive use. The big benefit for artists is the authoring environment, and in-engine results stay in lockstep. What the artist sees in a path-traced viewport is faithful to the final shipped product’s lighting, boosting confidence and reducing back-and-forth adjustments. This increased consistency means fewer surprises and less tweaking late in development, which in turn translates to shallower iteration cycles for teams, with engineers needing as many ad-hoc fixes for visual discrepancies and artists spending less time fighting technical limitations. Everyone works with the same view of the project, accelerating convergence toward the intended artistic vision.
Figure 8. Visualization of real-time path tracing at 1 SPP compared to Cycles’ Final Render (4096 SPP).
Conclusion
The Jungle Ruins scene provided a platform to explore and address significant technical challenges in real-time path tracing. By consolidating high-frequency detail rendering, large-scale animated instancing, and alpha testing into a unified workload, we revealed intricate interactions and practical performance bottlenecks specific to complex scenes. Our modular and scalable approach to scene design streamlined the workflow between art and research and allowed for targeted optimization and rapid iteration.
This focused exploration demonstrated the benefits of real-time path tracing in simplifying workflows, such as lighting and asset authoring, and ensuring a consistent result from content creation to final runtime. Despite the technical constraints and trade-offs, the strategic management of geometry, textures, and instancing allowed us to balance high-quality visuals with performance efficiency, highlighting practical techniques and considerations valuable to technical artists.
As real-time rendering continues to evolve, the insights gained from building the Jungle Ruins scene and embarking on this research exploration journey offer a strong foundation for future content creation and research direction, and contribute directly to the ongoing advancement of path-tracing applications in production environments.
Figure 9. Jungle Ruins scene.
Series
- Path Tracing a Trillion Triangles
- Neural Image Reconstruction for Real-Time Path Tracing
- Jungle Ruins Scene: Technical Art Meets Real-Time Path-Tracing Research - this post
- Path Tracing Massive Dynamic Geometry in Jungle Ruins
- Assessing Video Quality in Real-time Computer Graphics