DOOM: The Dark Ages pushes real-time graphics to new limits by integrating RTX neural rendering and path tracing, setting a new standard for how modern games balance visual ambition with fast, fluid gameplay. To explore the motivations, solutions, and lessons from overcoming these technical challenges, we spoke with Billy Khan, the director of engine technology at id Software.
Why did id Software decide to integrate path tracing into DOOM: The Dark Ages?
Path tracing is the next logical advancement in ray-traced visuals, and we’re eager to push the boundaries of lighting improvements in id Tech 8 (game engine used for DOOM: The Dark Ages). Demonstrating cutting-edge visual techniques in a commercially released game is crucial, as it holds more significance than showcasing them in a meticulously crafted, academic demo environment.
A new rendering technique is only truly viable if it can be integrated within the constraints of a shipping product, ensuring it works harmoniously with the existing rendering pipeline without introducing noticeable artifacts. Scalability across a wide range of hardware configurations is a core mission for us. We prioritize developing impactful features for both current and future high-end GPUs, while also ensuring the game performs optimally across a wide range of devices.
What are the benefits of path tracing compared to ray tracing?
Path tracing is more computationally expensive than ray tracing—it’s a level-up of current ray tracing techniques. The added costs deliver more realistic, physically accurate lighting and surface shading.
Dynamic RTX Global Illumination is more accurate with path tracing than with probe-based ray tracing solutions. Path tracing handles indirect emissive light contributions on surfaces better, produces more realistic, softer shadows, and generates more accurate reflections on rough surfaces. Overall, ray tracing provides fantastic results, but path tracing takes image quality and immersion to the next level.
What motivated your team to optimize GPU surface shading and integrate new rendering solutions?
For high-performance, visually complex scenes, optimizing both surface shading and transparency is crucial. Techniques that allow the GPU to intelligently reorder shading work and skip unnecessary calculations—like on pixels fully occluded by alpha-tested geometry—yield significant speedups. Both NVIDIA Opacity Micro-Map (OMM) and Shader Execution Reordering (SER) have been extremely useful for optimizing surface shading on the GPU. Some of our biggest optimizations for path tracing resulted from utilizing shader execution reordering.
What are the benefits from OMM and SER?
Opacity Micro-Map benefits any type of alpha-tested graphical workload—common examples include vegetation and particle systems. OMM reduces the shading rate by encoding the shading needs of micro-triangles onto larger triangles. This information is then used to short-circuit and avoid shading any pixels that are either fully transparent or opaque.
Shader Execution Reordering (SER), on the other hand, allows the GPU to optimize the order in which different types of shading work are performed, improving efficiency in scenes with complex lighting and materials. Both techniques are versatile tools for developers pushing the limits of real-time graphics.
How long did it take to implement path tracing?
From start to finish, it took about six months to implement and ship path tracing in DOOM: The Dark Ages.
Do you plan to integrate path tracing into future titles?
Absolutely. Path tracing is now part of id Tech 8. We’ll continue research and development in this area. As with all technology, we evaluate its use on a per-game basis to ensure it meets the needs of the game and delivers value to players. I believe path tracing will continue to evolve and will be used in more games in the coming years.
How has DLSS 4 changed your approach to image quality and game performance?
DLSS 4 brought in the new transformer model, which made significant improvements to super resolution and denoising that we were able to leverage in DOOM: The Dark Ages. Using the new transformer model with path tracing, Super Resolution and Ray Reconstruction can work in tandem, which not only improves the quality of upsampled pixels, but when compared to traditional denoisers, combining the two also improves performance. DLSS 4’s Multi Frame Generation enhances smoothness and can boost frame rates sufficiently to meet or surpass the maximum refresh rates of players’ display devices.
How valuable is DLSS Ray Reconstruction?
DLSS Ray Reconstruction added much-needed fidelity to denoising our path-traced results while preserving more detail. It’s a strong improvement in performance and quality over traditional denoising techniques. I’m excited to see further innovation here, which will empower more devices to play games with higher visual fidelity.
Can you explain why transformer-based upscaling models outperform older convolutional neural networks (CNNs)?
There are three main points.
- Transformer models better combine spatial and temporal data—motion vectors, depth information, and frame history—reducing ghosting and other artifacts in fast-moving games.
- Transformer models evaluate and model all pixels simultaneously, whereas CNNs are more restricted, usually processing only smaller regions. This allows transformers to better handle edges, leading to improved image quality.
- Transformer models go beyond simple pattern recognition, enabling more accurate reconstruction of surface details and shapes, which ultimately produces upsampled images with fewer artifacts.
Can you share any final tips or lessons learned for other developers looking to integrate AI and path tracing technologies?
Leveraging super resolution and denoisers that use AI-backed transformer models can significantly boost performance while preserving (or improving) image quality and resolution. I recommend starting with general ray tracing to understand its implications on your rendering pipelines. Optimal path tracing solutions are usually built on a robust ray tracing infrastructure, as it’s more GPU-demanding.
Keep your total shader count and complexity as low as possible for the best performance. Memory costs increase when using ray tracing and path tracing, since geometries must be stored in bounding volume hierarchies (BVH). It’s important to understand the cost differences between fully rebuilding BVH structures and refitting them—refits can be up to 10x faster. Understanding the relationship between your bottom-level acceleration structures (BLAS) and top-level acceleration structures (TLAS) is also key. Some of the highest costs for ray tracing and path tracing come from shading pixels—Shader Execution Reordering can help here, alongside traditional shading optimizations.
Get Started with Neural Rendering in Your Game
As seen in DOOM: The Dark Ages, leveraging innovations like neural rendering technologies empowers developers to push the boundaries of real-time graphics performance and fidelity—advancing both technology and the player experience.
See our full list of neural rendering and AI resources for game developers here.
