AMD’s DGF SuperCompression reduces asset size for future GPU support
AMD has released DGF SDK 1.2, adding a new feature called DGF SuperCompression. The update is aimed at reducing the storage footprint of geometry data used by Dense Geometry Format, AMD’s block-based compression format for dense meshes. This is a follow up to our previous coverage from last year.
Dense Geometry Format, or DGF, is designed for future GPU architectures with direct hardware support. AMD explains this as a hardware-friendly format for geometry compression, while the current SDK remains open source and supports all GPU vendors through DirectX 12 and Vulkan.
The new DGF SuperCompression, or DGFS, is not directly consumed by hardware. Instead, it acts as a smaller storage format for DGF data. AMD says DGFS can exactly reconstruct the original DGF blocks and can also decode to conventional vertex and index buffers, which allows the same assets to run on non-DGF hardware.
Source: AMD
According to AMD’s own test data, DGFS is roughly 30% smaller than raw DGF data in some examples. The Dragon model drops from 29.25MB to 20.15MB, while the Statuette model drops from 40.99MB to 29.31MB. When GDeflate compression is applied, AMD says DGFS remains roughly 20% smaller than DGF, with listed savings of up to 22.22%.
Source: AMD
AMD also published decode time results using a Ryzen 9 7950X system with 64GB of DDR5-6000 memory and a Radeon RX 9070 XT. A 10 million triangle Statuette model decoded to meshlets in 0.15 seconds, while DGF block decode took 0.22 seconds. AMD says the results are for a single CPU core and that a GPU-based decoder is also possible. So yes, it looks to be working on RDNA4 cards too, so it shouldn’t be limited to RDNA5 and later.
Rival to NVIDIA RTX Mega Geometry?
AMD DGF is similar but not directly compatible with NVIDIA RTX Mega Geometry. Both technologies address dense geometry in ray-traced rendering, but DGF is a geometry compression format, while RTX Mega Geometry focuses on clustered acceleration structure building. DGF SuperCompression can still decode to conventional mesh data, which allows content to run on GPUs without DGF hardware support.
Source: AMD GPU Open
