Microsoft rewrites ray tracing foundation: test scene frame rate soars 47x, VRAM cut by 90%. Microsoft has released a major preview update to the ray tracing pipeline in the DirectX developer blog, with the official engineer's exact words being "the most important update since DXR was born in 2018." The update ships with the DirectX 12 Agility SDK 1.722 preview, and its core consists of two new technologies: clusters

📜 Table of Contents

• The Numbers from Official Test Scenes
• Building Blocks and Partitions: Where the Savings Come From
• Microsoft Itself Pours Cold Water First
• Can Your GPU Use It?

Microsoft Rewrites Ray Tracing Foundation: Test Scene Frame Rate Soars 47x, VRAM Cut by 90%

Microsoft has released a major preview update to its ray tracing pipeline on the DirectX developer blog, with the official engineer's own words being "the most significant update since DXR was born in 2018." The update ships with the DirectX 12 Agility SDK 1.722 preview, and its core is two new technologies: Clustered Geometry and Partitioned Top-Level Acceleration Structure (PTLAS), targeting ray tracing's two biggest pain points: slow acceleration structure builds and heavy VRAM consumption.

The Numbers from Official Test Scenes

Microsoft ran a comparison using NVIDIA's Zorah demo scene: with Clustered Geometry enabled, frame rate jumped from 1.8 fps to 84 fps—roughly a 47x speedup; per-frame acceleration structure build time dropped from 482 ms to 3.06 ms; ray tracing itself went from 73.8 ms to 8.47 ms; and VRAM used by acceleration structures fell from 12.7 GB to 774 MB, a 93.9% reduction. Digital Foundry cited another test on the RX 9070 that's even more dramatic, with build time dropping 254x. PTLAS was tested separately on a domino collapse scene: a full rebuild costs 2.88 ms per frame, while a partitioned update takes just 0.79 ms.

Building Blocks and Partitions: Where the Savings Come From

The root of ray tracing's slowness lies in acceleration structures. Game visuals change every frame, yet ray tracing needs to build an index of the scene first—the traditional approach is roughly like redrawing the entire city's index every frame. Clustered Geometry breaks models into small blocks of at most 256 triangles, so when something changes, only the affected blocks get rebuilt; PTLAS divides the world into districts, so when a character walks through a block, only that block's index is recalculated. The two features can be used separately or together, and require no new hardware acceleration.

Microsoft Itself Pours Cold Water First

Under the official charts, there's a caveat: this is just a test scene—don't assume games will see gains of this magnitude. Real game scenes are far more complex, and the benefits will be somewhat smaller. There are also a few details the press releases don't mention: the new approach's temporary VRAM for builds actually increases; PTLAS trades build speedup for a slightly slower ray tracing step; engines need pipeline changes to integrate it, and the first batch of games isn't expected until next year. The closest technical precedent is NVIDIA's own RTX Mega Geometry, already in use in the PC version of Gears of War: E-Day—this Clustered Geometry is essentially its vendor-neutral counterpart.

Can Your GPU Use It?

Old NVIDIA card owners are luckier than old AMD card owners this time: NVIDIA included the 20-series in its list, while AMD's list has no RX 7000.

This whole thing is still a developer preview—regular gamers can't do anything with it for now except wait. Wait for engines to integrate it, wait for games to ship, wait for the first batch of titles next year to prove themselves. But this is already enough to get gamers excited. In an era of wildly inflating GPU prices, this is undoubtedly the best consolation for players.