AMD is doubling down on a future where games push far beyond today’s polygon budgets, and it’s giving a clearer look at how Dense Geometry Format (DGF) is meant to get there. The goal is straightforward: make ultra-detailed geometry practical in real-time ray tracing and other demanding 3D workloads, without ballooning memory use or triggering performance-hurting stalls.
Why geometry tech like DGF matters more than ever
Modern engines are rapidly raising the ceiling for scene complexity. Techniques that enable extremely detailed models often rely on lots of tiny triangles. That can work well for traditional rendering approaches, but ray tracing introduces a different set of challenges. If geometry data must be heavily decoded or converted before it can be used, it can add latency, increase memory pressure, and contribute to stuttering or inconsistent frame pacing. Even when acceleration structures are built with hardware support, the underlying data can still be too large and inefficient for the type of content developers want to ship in the next wave of games.
AMD’s view is that current ray tracing acceleration workflows behave too much like a “black box,” creating inefficiencies such as:
1) Memory allocations that must be sized for worst-case compression outcomes, which raises minimum memory footprints and complicates the build process.
2) Additional storage requirements to preserve exact triangle ordering for index references, which increases memory consumption.
3) Mandatory conversions into hardware-specific formats that can hurt performance, power, or silicon efficiency. This kind of forced runtime transcode also discourages denser designs that are harder to encode, indirectly pushing memory use even higher.
How AMD DGF works: meshlets and compact blocks
DGF is built around the idea of storing and streaming small geometry clusters instead of treating a full scene’s geometry as a monolithic chunk. In practice, AMD takes a standard triangle mesh and divides it into compact “meshlets.”
Each DGF meshlet includes:
– 64 vertices
– 64 triangles
– Stored inside a 128-byte DGF block (including metadata)
A group of these DGF blocks then represents the complete mesh. This structure is designed to scale better as geometric detail climbs, especially for real-time ray-traced rendering, but also for content creation, virtual production, and other real-time 3D use cases.
DGF SuperCompression: smaller storage today, bigger gains later
Alongside DGF, AMD also introduced a complementary step called DGF SuperCompression. Created in collaboration with Samsung and other software developers, this is positioned as a standardized, efficient geometry compression approach intended to reduce storage cost and make dense geometry more manageable.
Key point: DGF SuperCompression can reduce storage costs by up to around 30% by further compressing DGF data. The tradeoff is that SuperCompressed data is no longer directly consumable by hardware. However, AMD says it can be reconstructed exactly back into the original DGF blocks and can be decoded efficiently into conventional vertex and index buffers. That’s important because it means the content can still run on hardware that doesn’t fully support DGF natively.
AMD also notes a generational angle:
– Current Radeon hardware (including RDNA 4-class GPUs) can use DGF SuperCompression to get those storage reductions.
– Future architectures (often discussed in the context of what comes after RDNA 4) are expected to unlock more complete support and larger benefits.
Real-world examples: storage savings and decode performance
Using a Radeon RX 9070 XT, AMD shared sample results across several models (Crab, Dragon, Statuette, Buddha, Bike) showing that SuperCompression can meaningfully shrink geometry storage. Depending on the model, reported savings range from the high teens to just over 30%.
AMD also provided decode timing comparisons (meshlet decode vs. DGF decode) and listed the test system used, including a Ryzen 9 7950X, 64GB DDR5-6000, and Windows 11 (2025 update). The published decode times are small, but they still illustrate a real cost to decoding, reinforcing why AMD is pushing for more efficient end-to-end geometry formats as complexity increases.
How DGF fits into the next generation of ray-traced games
The industry is clearly heading toward much denser scenes, and upcoming AAA titles are already teasing that direction. At the same time, competing approaches are also emerging to solve similar problems, especially around ray tracing performance with extremely dense, dynamic geometry.
AMD’s DGF is meant to be a building block for future GPU designs and rendering techniques, including neural rendering directions the company has discussed for upcoming architectures. There’s no firm public timeline for when fully DGF-focused hardware arrives in consumer products, but AMD has been increasingly vocal about its roadmap features and long-term rendering plans, including work tied to its broader collaboration efforts with console partners.
Bottom line: DGF is AMD’s bet that the next leap in visual fidelity won’t just come from better lighting and ray tracing effects—it’ll come from making massive geometric detail efficient to store, stream, and render. If that bet pays off, the end result should be richer worlds, more detailed assets, and smoother performance in the ray-traced games that are clearly on the horizon.
