Microsoft announced DirectX Raytracing 1.2 (DXR 1.2) at GDC 2025, introducing two technologies that promise to significantly change how ray tracing performs on modern GPUs: Shader Execution Reordering (SER) and Opacity Micromaps (OMM). The announcement was made in partnership with AMD, Intel, NVIDIA, and Qualcomm, along with game studio Remedy Entertainment, whose game Alan Wake 2 served as the real-world showcase for the new features.
SER addresses one of the oldest and most persistent problems in ray tracing: shader divergence. When a GPU processes rays, each one can bounce off completely different surfaces, a glass window, a patch of grass, a metal wall, and each surface requires a different shader to run.
The problem is that GPUs are designed to work at maximum efficiency when all threads in a group, called a warp or wavefront, execute the same instructions simultaneously. When they don’t, some threads sit idle while others finish, and that wasted time translates directly into lost performance. SER introduces new HLSL primitives that allow the application itself to tell the GPU how to group and sort shader work, keeping threads coherent and eliminating much of that idle time. Microsoft claims performance gains of up to 2x in some scenarios as a result.
Opacity Micromaps work alongside SER to tackle a separate but related problem: alpha-tested geometry. Foliage, fences, hair, and other semi-transparent objects have always been extremely expensive in ray-traced scenes because every ray that hits them has to run an AnyHit shader to determine if it passed through or was blocked. OMMs pre-compute that opacity data at a sub-triangle level, so the GPU can skip those costly shader invocations in most cases. Microsoft claims up to 2.3x performance improvements in path-traced games with OMM enabled.
Real games are already showing the numbers
Remedy’s demo of Alan Wake 2 at GDC 2025 was the clearest proof that these gains aren’t just theoretical. In dense Pacific Northwest forest scenes with 37 million rays per frame, using SER and OMM together dropped ray tracing costs on an RTX 4090 from 16.8ms down to 10.2ms, roughly a 39% reduction in rendering time for one of the most visually demanding games available. Remedy’s CTO Mika Vehkala presented the results alongside Microsoft at the DirectX State of the Union session.
Outside of Alan Wake 2, the results are equally compelling in other implementations. In Unreal Engine 5, SER integration has produced up to 40% faster ray tracing operations in complex scenes with hardware ray-traced reflections and translucency. In the MachineGames path-traced lighting implementation for Indiana Jones, enabling SER alone dropped frame time from 4.08ms to 3.63ms, an 11% gain, and further optimization pushed that to 24%.
Microsoft also published a synthetic benchmark specifically designed to maximize divergence, where SER produced a 40% framerate increase on an RTX 4090 and a 90% framerate increase on Intel Arc B-Series GPUs. Those numbers represent an upper bound given the artificial conditions, but they illustrate the ceiling of what the technology can deliver when divergence is the primary bottleneck.
Which GPUs actually support it, and who’s still waiting
NVIDIA’s RTX 40 and 50 Series GPUs have full hardware acceleration for SER, which was purpose-built into the Ada Lovelace architecture. RTX 20 and 30 Series GPUs also support SER at the driver level, but the reorder step is effectively a no-op on those older architectures, meaning developers can write SER code that runs across all RTX hardware without fragmenting their codebase, even if only newer cards see the actual performance benefit.
Intel confirmed support for SER on Arc B-Series Graphics and Core Ultra Series 2 processors via a developer preview driver, and those 90% gains in the synthetic benchmark suggest Intel’s architecture responds particularly well to the feature.
AMD is currently the notable gap. Current RDNA 2, 3, and 4 hardware does not perform actual thread reordering, AMD’s driver technically accepts SER code as required by Shader Model 6.9, but the MaybeReorderThread call does nothing on present hardware. Real hardware-accelerated SER on Radeon GPUs is expected to arrive with RDNA 5. Microsoft has stated AMD is working with them on broader adoption of DXR 1.2 technologies, so it’s a matter of when, not if.
For developers, SER being a required feature in Shader Model 6.9 means they can write for it without worrying about breaking things on unsupported hardware, unsupported devices simply ignore the reorder call with no performance penalty. What remains now is adoption: developers need to implement SER and OMM in their engines and titles, which takes time. But with Alan Wake 2 already demonstrating results, Unreal Engine 5 integration underway, and Shader Model 6.9 now officially out of preview, the foundation is in place for ray tracing to become meaningfully faster across a wide range of games in the near future.
So tell us, are you sitting on an RTX 40 or 50 series card ready to see these gains in action, are you on an older RTX praying the no-op doesn’t hurt you, or are you on Team Red watching from the sidelines waiting for RDNA 5? Drop your thoughts in the comments, we genuinely want to know where you stand on this one.