Ray tracing graphics, alongside upscaling and similar “more FPS” tricks, has been the main trend in games in recent years. Although ray tracing is commonly associated with Nvidia’s RTX marketing brand, it’s actually built on the DirectX API: DirectX Ray Tracing, or DXR, which launched before GeForce RTX. This technology is now getting a major upgrade to version DXR 1.2, which will bring new capabilities as well as better performance.
DirectX Ray Tracing reached version 1.1 in 2019, when features like inline ray tracing, the ExecuteIndirect capability, and other improvements were added. The new DXR 1.2 version therefore arrives after a significant gap. Microsoft announced it last month during the GDC 2025 conference. Like DXR 1.1 before it, DXR 1.2 too is not a radical shift that changes the fundamentals of how the GPU is used. Instead, it expands on the previous standard by adding support for new capabilities that have emerged in GPU hardware in the meantime. There are two key features in particular.
Shader Execution Reordering
The first is Shader Execution Reordering (SER). You may remember this term from the launch of GeForce RTX 4000 graphics cards, as it was one of the new features announced with the Ada Lovelace GPU architecture. SER allows the GPU to reorganize the order of shader execution to a certain degree, enabling more efficient use of computing resources.
The intended result of using SER is increased performance – its purpose is to squeeze more FPS out of the same hardware, and along with other optimizations, provide more headroom for demanding ray-traced graphics. However, this feature is not the same as out-of-order execution known from CPUs; it’s a simpler capability that requires software cooperation.
Nvidia is expected to support SER on GeForce RTX 4000 and also on RTX 5000 cards, on which should work better (on the other hand, support on older GPUs is unlikely). The company indicates that this technology is designed for specific scenarios and is not universally beneficial and therefore not enabled by default. Its use is based on an “opt-in” principle, meaning developers can enable it for particular shaders where profiling shows performance gains from SER. It’s still unclear whether the feature will work the same way in DXR 1.2 or whether the standardized version will aim for broader, default usage. Given the lack of support on older GPUs, it seems unlikely that games will use it implicitly (and require it to run) in the near future, anyway.
Opacity Micromaps
Another new feature aimed at improving ray tracing performance added by DXR 1.2 is Opacity Micromaps (OMM). This is an optimization in handling geometry during ray tracing – specifically in dealing with transparency and opacity of objects and their parts. OMM is meant to store data about the transparency of individual triangles and models more efficiently.
Micromaps replace alpha textures in this role and reduce the workload shaders have to perform by simplifying the evaluation of how light rays interact with objects. This should enable faster rendering of, for example, complex vegetation and similarly complicated game objects.
This feature was also previously introduced by Nvidia, showcased as part of the Ada Lovelace architecture. We therefore assume that OMM will once again be supported in GeForce RTX 4000 and newer GPUs, while RTX 2000 and 3000 generations likely won’t support it.
Support in GPUs from various manufacturers
These DXR 1.2 enhancements should eventually be supported across all GPUs beyond just the newer GeForce cards (if they don’t already support these features). Microsoft presented the DXR update alongside AMD, Intel, Nvidia, and Qualcomm (which makes GPUs used in Windows devices through their ARM processors), so the entire PC gaming ecosystem should be onboard. That said, initial support may vary between vendors, and some features might only appear in future GPU architectures.
According to Qualcomm, SER and OMM support will come to the company’s integrated GPUs only with the next generation of processors. Intel’s existing Arc GPUs already support SER, but not OMM – support for OMM is expected in a future generation of Arc graphics cards. We do not yet have precise information on the state of SER and OMM support in AMD Radeon GPUs.
Alongside these two DXR 1.2 updates, another change is coming to DirectX: a technology called Cooperative Vectors, which we’ve covered previously. This tech allows standard shader programs (including basic pixel shaders) to be combined with AI model inference (using smaller neural networks) within the same code.
This is particularly significant for GPUs with dedicated AI units, such as Nvidia’s tensor core-equipped GPUs. Previously, integrating the two was not straightforward, and Nvidia also promotes this as one of the key new features of its newest GPU architecture – this year’s Blackwell release. Intel has also confirmed that both generations of the company’s Arc GPUs – Alchemist and Battlemage (which also have specialized XMX units for neural network acceleration) – will receive driver support for Cooperative Vectors.
Source: Microsoft, VideoCardz
English translation and edit by Jozef Dudáš
⠀
