Microsoft announced this week its DirectX Raytracing (DXR) 1.2 application programming interface, which introduces new features that significantly improve visual quality and rendering performance by up to 2.3 times. AMD, Intel, Nvidia, and Qualcomm, along with game developers like Remedy, are working to integrate DXR 1.2 technologies into future gaming hardware and software.
The DirectX Raytracing 1.2 update includes Opacity Micromaps (OMM) and Shader Execution Reordering (SER), two technologies that boost performance in raytraced games by 2 (SER) to 2.3 times (OMM). Both technologies must be implemented into actual games or game engines to experience the performance benefits.
2X – 2.3X performance boost
One of the main issues with alpha-tested geometry (foliage, fences, hair, etc.) in raytracing is extra calculations required for light to determine whether it hits a surface or passes through. Opacity Micromaps (OMM) improve how alpha-tested geometry processes by applying a texture with an alpha channel to a flat surface. It then removes pixels below a certain transparency threshold. OMM reduces the number of times shaders need to be used, leading to higher efficiency and performance.
In the best-case scenario, Microsoft claims an improvement of 2.3 times. However, keep in mind that not all games and scenes contain a lot of elements like foliage and fences. For example, while S.T.A.L.K.E.R. 2 has loads of grass, leaves, and fences in practically all scenes, Cyberpunk 2077 barely has any foliage.
Shader Execution Reordering (SER) seems to be a more universal feature as it reorders how shaders are executed to avoid shader divergence. Shader divergence occurs when nearby pixels require shaders to do different tasks, a common situation in scenes with heavy raytracing effects, such as complex lighting, realistic shadows, and detailed reflections.
GPUs process shaders in parallel threads organized into groups called warps or wavefronts. Ideally, all threads within a group execute identical instructions simultaneously, maximizing GPU efficiency. Shader divergence occurs when threads in the same warp or wavefront need to perform different instructions. In this case, simultaneous execution is impossible, forcing the GPU to handle each instruction path separately, leaving some threads idle and increasing latency.
According to Microsoft, SER sorts or batches similar shader workloads together, reducing divergence and maximizing GPU utilization and speeding up rendering by up to two times.
Hardware support
Regarding hardware support, the situation is a mixed bag, which is common with new API features.
All Nvidia GPUs dating back to Turing (GeForce RTX 20-series) support Opacity Micromaps (OMM), so these graphics cards can potentially experience a performance boost once game developers implement them into their titles. Intel said its next-generation Celestial (Xe3) GPUs will also support OMM.
Nvidia's GPUs have supported Shader Execution Reordering (SER), starting with the GeForce RTX 40-series Ada Lovelace family. Intel said it looks forward to supporting SER "when it is available in a future Agility SDK.' However, whether it will be supported on Intel's Arc 'Alchemist' or Arc 'Battlemage' GPUs (or both) is unclear.
AMD does not seem to support OMM or SER on its RDNA 2/3/4 GPUs, though Microsoft said that the red company is working with it on the widespread adoption of these technologies. Also, AMD has certain scheduling optimizations that may mimic how SER works, so if game developers take time to optimize for Radeon GPUs, the latter may get some speed improvements.
Qualcomm also does not support OMM or SER, but it said it would on its next-generation integrated GPUs.
The preview version of DXR 1.2 will launch in April 2025.
