AMD DGF hardware decode is reserved for next-gen RDNA, here’s how it works

AMD blog post focuses on upcoming tech.

AMD’s latest GPUOpen post adds animation support to its Dense Geometry Format (DGF). You can use DGF on today’s GPUs through shaders, but true hardware-accelerated DGF is planned for “future RDNA GPUs”, which means RDNA5. DGF was first shown in 2024, and AMD also published a provisional Vulkan extension this summer.

What DGF does: it packs small chunks of a mesh (up to 64 vertices and 64 triangles) into 128-byte blocks. Each block stores positions in a compact way and uses a light topology scheme. The new sample shows how to animate by blending vertices on the GPU, then writing the updated positions back into those blocks.

Source: AMD

According to AMD, the ‘cost’ is low. On an RX 7900 XT, the blog shows animation and the re-packing step at under 1% of frame time. Most time still goes to building the BVH and tracing rays. The win is lower memory use and better cache behavior while moving or deforming models.

Technically all GPUs can run DGF via shaders today, but AMD is lining up hardware help for RDNA5. Reddit discussion, which I must say does a lot better job explaining what this technology is about, points out that NVIDIA has offered related geometry compression/cluster tech since RTX 40, and expanded it with RTX 50. Meanwhile, RDNA5 desktop cards are expected around late 2026, and the same tech is likely to show up in next-gen consoles.

Sources: AMD GPUOpen , Reddit