Microsoft has previewed two DirectX ray-tracing features that, in its own test, cut memory used by ray-tracing structures by about 90%. The bigger promise is a shared approach across graphics vendors, though developers will need to put it into games before players see the benefit.
Watch Desk analysis
What happened
TechSpot reports that Clustered Geometry breaks models into reusable chunks of up to 256 triangles, while Partitioned TLAS divides a scene into regions so only areas that have changed need updating. The aim is to avoid rebuilding large amounts of ray-tracing data from scratch whenever objects move or change detail.
On a Radeon RX 9070 test, Microsoft’s sample rose from 1.2 to 79 frames per second. Structure-building time fell from 788 milliseconds to just over three, while memory for those structures dropped from roughly 7.5GB to 774MB. Those are results from Microsoft’s sample, not a forecast for games: Microsoft warns that real-world gains should be nowhere near as large. The memory figure covers ray-tracing structures alone, not a game’s total VRAM use.
The features remain in preview and require Windows Developer Mode. TechSpot says Nvidia supports them on RTX cards, Intel on Arc B-series GPUs and recent Core Ultra chips, and AMD on Radeon RX 9000 cards so far. Developers still have to adopt them, and clustered builds require substantially more temporary scratch memory.
Why it matters
Ray tracing can demand a hefty memory budget, particularly when scenes change frequently. Reusing geometry data and updating only the parts of a scene that have changed could make more detailed ray tracing practical within existing hardware limits. This is a concrete graphics-memory development, not a new GPU generation dressed up in a clever name.
The cross-vendor DirectX route is notable too: developers could target one API rather than separate implementations. But preview support is not the same as a game players can install today, and a spectacular Microsoft sample is a starting point for testing, not a promise of 79 frames per second on your favourite title.
Our read
This looks like meaningful groundwork for better use of graphics memory, with the useful twist that it does not require new hardware on the supported cards listed. The sensible verdict is “promising, not yet a shopping list”: wait for games and independent testing before treating the sample numbers as your next upgrade plan.
What to watch
- When DirectX moves beyond preview and what hardware remains supported.
- Which games adopt Clustered Geometry or Partitioned TLAS.
- Whether independent tests show useful memory savings in real games.
- How developers manage the extra temporary scratch memory.
Discussion spark: Would you rather see developers spend freed graphics memory on richer ray tracing, or use it to improve performance on current settings?
Sources and evidence
- Source update (31 October 2026, 17:45 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.