The End of the 200GB Game? How Geometry Compression is Saving Your SSD
We’ve all been there: you go to download a new AAA title, only to realize it requires 150GB of free space. As game worlds grow more ambitious and textures become more lifelike, the sheer volume of data is hitting a wall. The bottleneck isn’t just how prompt your SSD can read data, but how much of that data is actually necessary to store.
Enter AMD’s Dense Geometry Format (DGF) and its latest evolution, DGF SuperCompression (DGFS). This isn’t just a minor tweak; it’s a fundamental shift in how 3D objects are stored and streamed into your GPU. By attacking the “geometry bloat,” AMD is paving the way for hyper-detailed worlds that don’t require a dedicated hard drive for every single game.
Decoding the Tech: DGF vs. DGFS
To understand where we’re going, we have to understand the difference between the format and the compression. DGF is a block-based format that packs tiny mesh pieces—up to 64 vertices and 64 triangles—into compact 128-byte blocks. This allows for a massive increase in geometric detail, effectively surpassing traditional raster methods like tessellation.
DGFS (SuperCompression), however, is the “wrapper” around that data. Think of DGF as the organized folder and DGFS as the ZIP file. DGFS doesn’t live in your GPU’s memory; it sits on your SSD or HDD. When the game needs that geometry, a CPU-based decoding process unzips it in real-time during asset streaming.
According to recent data, when combined with GDeflate, DGFS achieves an average 22% reduction in compressed game asset streams. In an era where games are pushing 200GB, a 22% saving is the difference between fitting three games on a drive or only two.
The Hardware Horizon: RDNA 4 and Beyond
One of the most intriguing aspects of this rollout is the hardware compatibility. While designed for future architectures, testing on the Radeon RX 9070 XT suggests that RDNA 4 will likely support the DGF graphics stack. This means users won’t have to wait for a generational leap to RDNA 5 or UDNA IP to see the benefits of denser, more efficient worlds.
The Battle for Geometry Supremacy: AMD vs. NVIDIA
This move places AMD in direct competition with NVIDIA’s RTX Mega Geometry. Both companies are racing toward a future where “polygons” as we know them are replaced by “dense meshes.” The goal is to allow artists to create cinematic-quality assets without the manual labor of creating multiple Levels of Detail (LODs).
However, AMD is taking a different strategic path: Open Standards. By partnering with Samsung to develop a multivendor DGF extension for Vulkan, AMD is attempting to make dense geometry a universal language rather than a proprietary wall. If this succeeds, we could see a standardized way for all GPUs to handle hyper-dense geometry, accelerating the adoption of the tech across the entire industry.
Future Trend: The Death of the Load Screen
The synergy between DGFS and real-time CPU decoding points toward a future of “seamless streaming.” When geometry is this compressed, the “pipe” from the SSD to the GPU becomes less congested. We are moving toward a world where the environment evolves in detail as you move through it, with no perceptible loading or “pop-in,” because the data is small enough to be decoded and deployed instantly.
Frequently Asked Questions
No. AMD has developed a CPU-based decoding process designed to happen in real-time during asset streaming. On a Ryzen 9 7950X, a 10-million triangle model decoded in as little as 0.15 seconds.
Not necessarily. While future GPUs will have direct hardware support, the current SDK allows DGFS to decode into conventional vertex and index buffers, meaning it can run on non-DGF hardware.
Texture compression handles the “skin” of an object (the image). DGF and DGFS handle the “skeleton” and “muscle” (the actual 3D shape and vertices).
What do you think? Will geometry compression be enough to stop the endless growth of game install sizes, or are we just delaying the inevitable? Let us know in the comments below, or subscribe to our newsletter for the latest deep dives into GPU architecture!
