AMD’s open-source Dense Geometry Format (DGF) compresses geometry data by up to 31%, dramatically reducing storage costs and accelerating ray tracing. With the new SuperCompression layer, a Samsung partnership for mobile, and RDNA 5 hardware optimization on the horizon, DGF is positioned to reshape how games handle path tracing.
AMD released an updated Dense Geometry Format (DGF) SDK on May 7, 2026, introducing DGF SuperCompression (DGFS), a technology that shrinks raw geometry data by up to 31.09% and compressed on-disk data by roughly 22%. DGF is a block-based, open-source geometry compression standard designed to close the gap between increasingly complex game geometry and the demands of real-time ray tracing. Unlike proprietary solutions, DGF works across vendors through Vulkan support and is licensed permissively for anyone to implement.
What Problem Does DGF Solve?
Modern game engines like Unreal Engine 5 with Nanite have pushed triangle counts into the tens of millions per scene. These technologies use compact triangle formats with software rasterization via compute shaders, which creates a serious bottleneck for ray tracing. Current ray tracing APIs treat acceleration structure builds as a “black box,” requiring geometry to be decoded from custom formats into uncompressed data before the driver converts it into hardware-specific BVH structures. This pipeline inflates memory usage, adds latency, and can cause stuttering frame rates.
According to AMD’s GPUOpen documentation, the current API design has three fundamental limitations: pre-build memory must be sized for worst-case compression, implementations must store enough data to reproduce exact triangle ordering, and mandatory runtime transcoding discourages denser encoding approaches. DGF addresses all three by moving geometry compression into a standardized hardware-native format, doing for geometry what DXT, ETC, and ASTC did for texture compression.
How DGF Works at the Block Level
DGF packs triangle meshes into 128-byte cache-aligned blocks. Each block holds up to 64 vertices and 64 triangles along with metadata. This structure ensures that all data for any individual triangle can be retrieved in a single memory transaction, a property AMD describes as essential for efficient hardware ray traversal and highly desirable for rasterization.
The compression process divides standard triangle meshes into smaller meshlets, then encodes them into DGF blocks. Developers control the tradeoff between accuracy and memory through a configurable target bit-width parameter. At 16 bits, a test model compressed to 4.4 MB; at 12 bits, the same model dropped to 3.0 MB. Importantly, DGF preserves mesh connectivity and topology, so there is no need for labour-intensive remeshing or resampling.
DGF SuperCompression: Smaller Files, Broader Compatibility
While DGF blocks are optimized for hardware consumption, they are not ideal for storage because vertex positions and compression parameters must be duplicated across blocks. SuperCompression (DGFS) solves this with cluster-level de-duplication and delta encoding. AMD’s benchmarks using a Radeon RX 9070 XT show significant savings across test models:
| Model | Triangles | DGF Size | DGFS Size | Savings |
|---|---|---|---|---|
| Dragon | 7.22M | 29.25 MB | 20.15 MB | 31.09% |
| Statuette | 10.00M | 40.99 MB | 29.31 MB | 28.48% |
| Crab | 2.14M | 10.22 MB | 8.48 MB | 17.06% |
| Buddha | 1.09M | 4.94 MB | 3.95 MB | 20.03% |
| Bike | 1.68M | 6.96 MB | 5.54 MB | 20.47% |
When GDeflate compression is applied on top (the realistic gaming scenario where DGFS data does not persist in memory), DGFS still provides roughly 20% additional savings over compressed DGF blocks. Decode times are fast enough for real-time streaming: a 10-million-triangle model decodes on a single CPU core in 0.15 seconds to meshlet format and 0.22 seconds to DGF blocks.
A crucial feature of DGFS is dual-target decoding. The same DGFS stream can decode either to DGF blocks (for hardware with native DGF support) or to a conventional vertex and index buffer (for non-DGF hardware). This means developers ship one asset format that works everywhere.
How DGF Compares to NVIDIA RTX Mega Geometry
DGF is frequently compared to NVIDIA’s RTX Mega Geometry, but the two technologies attack different points in the rendering pipeline. As Digital Foundry’s analysis noted, they are complementary rather than competing. RTX Mega Geometry focuses on cluster-level acceleration structures (CLAS) and batch BVH updates, enabling granular, GPU-driven scene management. On an RTX 5090, Mega Geometry achieves over a terabyte per second of throughput for BVH builds, translating to 30 to 40 billion triangles per second.
DGF, by contrast, compresses the geometry data itself before it enters the BVH build pipeline. AMD’s GPUOpen team explicitly states that “DGF is not a response to the development of cluster-level acceleration structures, but an orthogonal, complementary technology.” In theory, a future architecture could use both compressed geometry input (DGF) and cluster-level acceleration management (CLAS-style) simultaneously for maximum efficiency.
NVIDIA’s RTX Mega Geometry is currently shipping in hardware on RTX 50 series GPUs and has been deployed in Alan Wake 2. A new foliage system built on Mega Geometry was announced for The Witcher 4 at GDC 2026. DGF is currently software-emulated, with native hardware support coming with RDNA 5.
Samsung Partnership and the Mobile Roadmap
AMD announced a partnership with Samsung to develop a multivendor DGF extension for Vulkan that both companies intend to implement. Gabe Dagani, Senior Director of GPU Software at Samsung, confirmed that the companies are working to bring “advancements such as the DGF extension to future mobile devices.” Samsung has used AMD’s RDNA architecture in its Exynos processors since 2022, when it first brought ray tracing to smartphones.
The mobile angle is strategically important. Mobile GPUs operate under tight memory bandwidth and power constraints, exactly the conditions where geometry compression delivers outsized benefits. If DGF becomes a cross-platform standard supported by both desktop and mobile GPU vendors, it could unify geometry handling across the entire rendering ecosystem.
RDNA 5, Xbox Project Helix, and the Console Generation
AMD has confirmed that DGF is optimized for the upcoming RDNA 5 architecture, which will power both next-generation Radeon desktop GPUs and the custom SoC inside Microsoft’s Xbox Project Helix. At GDC 2026, Microsoft revealed that Project Helix will deliver a 10x improvement in ray tracing performance over current-generation consoles, with full path tracing capability.
RDNA 5 introduces dedicated hardware blocks for ray tracing (called Radiance Cores), neural rendering arrays, and universal compression. DGF fits directly into this architecture: instead of software-emulated decompression, future GPUs will natively consume DGF-compressed geometry during BVH builds, dramatically reducing both build time and memory footprint.
AMD also unveiled FSR Diamond alongside Project Helix. This next-generation image enhancement suite includes ML-based upscaling, multi-frame generation, and ray regeneration for both ray tracing and path tracing. FSR Diamond is reportedly exclusive to RDNA 5 hardware, though AMD has not officially confirmed PC GPU availability details. The combination of DGF for geometry efficiency and FSR Diamond for frame quality positions RDNA 5 as a significant architectural leap for real-time path tracing.
Open Source and the SDK
The DGF SDK (version 1.2) is fully open-source on GitHub under a permissive license. It includes CPU libraries for compression and decompression, optimized HLSL code for GPU-based decompression, a reference clustering pipeline compatible with MeshOptimizer, and a command-line encoder with an interactive viewer supporting OBJ and PLY models. The SDK works with DirectX 12 and Vulkan, supports SPIR-V and DXIL shader intermediate representations, and runs on GPUs from all vendors.
AMD has also released a provisional Vulkan extension (VK_AMDX_dense_geometry_format) that allows DGF data to be fed directly to acceleration structure builds. Driver support is available in AMD Software: Adrenalin Edition drivers. On hardware with native DGF support in the future, this extension will significantly improve BLAS build time and reduce BLAS memory footprint.
What This Means for Gamers
In practical terms, DGF and DGFS will enable game developers to create denser, more detailed open-world environments without proportionally increasing storage requirements or tanking frame rates. The 20 to 31% reduction in geometry data size translates to smaller download sizes, lower VRAM consumption, and faster scene loading, all of which directly benefit players.
For path tracing specifically, geometry has been one of the biggest bottlenecks. Dense scenes with millions of polygons require massive BVH structures that eat into the memory and compute budget needed for lighting calculations. By compressing geometry at the format level, DGF frees up resources for more complex lighting, more detailed environments, and smoother performance across a wider range of hardware.
As both AMD (through DGF and RDNA 5) and NVIDIA (through RTX Mega Geometry and DLSS 4.5) push geometry and rendering efficiency forward, the 2027 console generation is shaping up to be the first where full path tracing is a realistic default rather than a high-end luxury. For PC gamers, these technologies should trickle into titles well before next-gen consoles arrive, particularly in games built on engines that adopt DGF early.
Key Questions About AMD DGF
Does DGF only work on AMD GPUs?
No. The DGF SDK is open-source and vendor-agnostic. Software-based encoding and decoding runs on any hardware. The Vulkan extension is currently AMD-specific, but the Samsung partnership aims to create a multivendor standard. Native hardware acceleration will debut with RDNA 5, but DGFS can decode to standard vertex buffers for non-DGF devices today.
Is DGF lossy or lossless?
DGF compression is lossy at the initial quantization step: vertex coordinates are quantized to a target bit-width (configurable from 12 to 16+ bits). After that step, DGFS exactly reconstructs the DGF blocks with no additional loss. Developers can fine-tune the precision to balance quality against compression ratio.
Will DGF reduce game download sizes?
Geometry is one component of overall game size alongside textures, audio, and video. For geometry-heavy open-world titles, DGFS can reduce the geometry portion by 20 to 31%. The effect on total download size depends on how much of a game’s data is geometry, but the savings are meaningful for titles pushing modern polygon counts.
When will games start using DGF?
The SDK is available now for engine developers to integrate. Software-emulated DGF can run on current hardware. Full hardware-accelerated support will arrive with RDNA 5 GPUs and next-generation consoles expected from 2027 onward. Games targeting those platforms are likely already evaluating DGF integration.
AMD’s DGF represents a foundational shift in how game geometry is stored, transmitted, and processed. Its open-source nature, vendor-agnostic design, and proven compression results make it a strong candidate to become the industry’s standard geometry compression format. Combined with RDNA 5’s hardware support and the broader push toward real-time path tracing from both AMD and NVIDIA, DGF is one of the key technologies that will define the visual quality of games over the next hardware generation.
