Microsoft presented a comprehensive update to its DirectX toolchain at the Game Developers Conference (GDC) 2025. The focus: DirectX Raytracing 1.2 (DXR 1.2), Shader Model 6.9 with “Cooperative Vectors”, as well as neural rendering technologies, which are to be implemented on a broad hardware basis in collaboration with NVIDIA, AMD and Intel. The presentation turned out to be less of a marketing firework display and more of a technical feasibility study with potential added value – at some point.
DXR 1.2: More structure, less shader overhead
With version 1.2, Microsoft is introducing two new core functions: Opacity Micromaps (OMM) and Shader Execution Reordering (SER). OMM is intended to make the display of alpha-tested geometry drastically more efficient by describing the transparency of textures in a much more granular way and thus reducing unnecessary shader calls. According to Microsoft, this brings – in path-traced scenes – performance gains of up to 2.3 times. A value that is more likely to be achieved under ideal conditions in the lab than on an average gaming PC. SER, on the other hand, ensures dynamic reordering of shader executions in order to reduce the divergence within the GPU pipeline. The goal: more parallelization, less waiting times, better utilized compute units. Here, too, there is talk of a performance increase of up to 2x – although this is highly scene-dependent. According to Microsoft, NVIDIA has already included both features in its drivers, while AMD, Intel and even Qualcomm are still working on integration. A rapid market penetration is therefore not to be expected.
Cooperative vectors: Shader Model 6.9 thinks outside the box
The concept of so-called “cooperative vectors” is more exciting – because it is more significant in the long term. In Shader Model 6.9, it will be possible to execute vector and matrix operations hardware-accelerated via new engines within the GPU. Microsoft thus promises a deeper anchoring of AI-supported rendering directly in the graphics pipeline. At the GDC, Intel, for example, demonstrated a 10-fold acceleration in the use of neural compression methods – specifically in the so-called Neural Block Texture Compression, which is supposed to deliver comparable image quality with lower memory requirements. This is particularly interesting for mobile GPUs and APU-based systems, where bandwidth is often a limiting resource.
NVIDIA, AMD, Intel – all in the same boat, but rowing differently
While NVIDIA already provides a developer toolkit in the form of the Neural Shading SDK, which is DirectX-compatible and supports cooperative vectors, AMD and Intel are keeping a rather low profile when it comes to concrete software solutions. Nevertheless, all three major GPU providers have signaled their support for neural rendering – which initially costs little, but in practice requires a long path through drivers, middleware and engines.
WARP gets an update – software rendering on steroids
In addition to GPU-side features, Microsoft has also updated its CPU-based rendering solution WARP (Windows Advanced Rasterization Platform). WARP sees itself as a reference platform for DX12 Ultimate and will also support ray tracing, mesh shaders and work graphs in future – completely without a GPU. For developers, this is a helpful option for troubleshooting, for the rest of the world it is more of a stopgap for defective hardware.
PIX, Agility SDK & tech demos: Everything is coming – but first as a preview
The new technologies will be available in the preview Agility SDK from the end of April 2025. Microsoft’s debugging and profiling tools (PIX) will also receive a corresponding update to fully support DXR 1.2. At the same time, the first tech demos were shown, including a revised version of Alan Wake 2, in which the performance gains of SER and OMM were illustrated – at least on slides.#
Many new paths, but no destination in sight yet
DirectX Raytracing 1.2 brings a lot of interesting approaches to make raytracing and AI-supported rendering more efficient and practicable. However, the reality remains sober: there are at least two years, several driver versions and a handful of half-baked early-adopter titles between specification and widespread availability. Anyone hoping for DXR 1.2 today will only get one thing for the time being: new abbreviations to memorize.
Source: Microsoft
16 Antworten
Kommentar
Lade neue Kommentare
Mitglied
Urgestein
Veteran
Urgestein
Mitglied
Urgestein
Veteran
Urgestein
Veteran
Urgestein
Urgestein
Urgestein
Veteran
Urgestein
Urgestein
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →