AMD has revealed a new ray-tracing research technique that could dramatically reduce how much video memory complex animated scenes require. In one demonstration, a workload that would normally demand more than 80 GB of VRAM reportedly used just 1.7 GB.
The technique is known as the tetrahedral cage approach, and AMD believes it could provide developers with a much more memory-efficient way of handling animated geometry in ray-traced environments.
While it remains a research project for now, the results could be particularly interesting for future games packed with vegetation, large environments and animated objects.
AMD Targets One of Ray Tracing’s Biggest Problems
Ray tracing has become increasingly common in modern games, but it comes with substantial hardware requirements. GPU performance is one part of the problem, while memory consumption can become another major limitation.
Animated geometry makes matters even more complicated.
Traditional clustered vertex-animation techniques give developers precise control over individual vertices. However, storing all of that animation information can consume enormous amounts of memory when scenes become sufficiently complex.
AMD’s researchers are experimenting with another approach.
Instead of storing everything using the traditional method, animation data can be represented using tetrahedral cages. These cages provide a more compact representation of the movement while substantially reducing the amount of information that needs to remain in GPU memory.
The trade-off is reduced per-vertex control, meaning it won’t necessarily replace existing animation techniques everywhere.
From More Than 80 GB to Around 1.7 GB
The numbers from AMD’s demonstration are what make the technology particularly interesting.
AMD demonstrated a densely populated environment containing grass and trees while using ray tracing. The test was performed on a Radeon RX 9070 XT equipped with 16 GB of GDDR6 memory.
Despite the amount of animated vegetation on screen, the demonstration reportedly ran above 60 FPS at 1080p.
More importantly, it consumed approximately 1.7 GB of video memory.
AMD estimates that recreating the same workload using its traditional clustered vertex-animation approach would require more than 80 GB of VRAM.
That represents a potential reduction of more than 40 times for this particular workload.
It doesn’t mean every ray-traced game can suddenly reduce its VRAM requirements by 40×. The comparison applies specifically to the animation technique demonstrated by AMD, and real games contain many other assets competing for GPU memory.
Still, the difference shows why AMD is researching the technology.
Foliage Could Be an Ideal Use Case
Not every object needs extremely precise vertex animation.
Vegetation is an obvious example.
A forest might contain thousands of individual trees, bushes and patches of grass moving in the wind. Players usually don’t need every blade of grass to use an expensive animation system with extremely precise vertex control.
Tetrahedral cages could therefore be particularly useful for large amounts of foliage.
The same principle could potentially apply to distant characters and other animated objects where small reductions in animation precision would be difficult to notice during normal gameplay.
Developers could then reserve more expensive animation techniques for objects closer to the player or situations where additional precision actually matters.
VRAM Is Becoming Increasingly Important
Modern games are already pushing graphics cards with 8 GB and even 12 GB of VRAM harder than they did several years ago.
Higher-resolution textures, larger environments, ray tracing and increasingly complicated geometry all compete for the same memory pool.
Simply throwing more memory at the problem isn’t always practical either.
Reducing memory requirements through smarter rendering techniques could therefore become just as important as increasing raw GPU performance.
If something similar to AMD’s tetrahedral cage system eventually reaches production games, developers could potentially build more complicated scenes without requiring enormous amounts of additional VRAM.
That could be especially valuable for lower-end GPUs and older hardware.
Consoles Could Eventually Benefit Too
AMD’s graphics technology extends far beyond desktop Radeon cards.
Current-generation consoles already rely heavily on AMD hardware, while future gaming systems are expected to continue using technologies developed by the company.
That makes research into more efficient ray tracing particularly interesting. Consoles operate within strict memory and performance limits, meaning techniques that significantly reduce memory consumption can give developers additional room elsewhere.
However, AMD has not announced that tetrahedral cages will appear in any upcoming console or GPU architecture.
For now, this remains a research project rather than a confirmed feature.
Impressive Technology, but Don’t Expect It Tomorrow
The demonstration certainly produces impressive numbers, but there is an important distinction between a research demo and technology being used inside commercially released games.
Developers would need appropriate tools, engine support and practical reasons to adopt the technique. Its limitations also mean traditional clustered animation isn’t suddenly obsolete.
Instead, tetrahedral cages could become another tool developers use depending on the scene.
For large amounts of foliage, background animation and other geometry where perfect vertex-level control isn’t necessary, the memory savings could make the trade-off worthwhile.
Going from more than 80 GB to roughly 1.7 GB in AMD’s demonstration shows just how much room there could still be for optimization.
Whether gamers actually see those benefits in future games will depend on what happens with the technology after the research stage.
