AMD’s next-generation GPU architecture is increasingly looking like it will be called RDNA 5, not UDNA—and a fresh leak adds more weight to that idea.
Recent updates spotted in the LLVM compiler stack show early enablement for “GFX13,” an important internal identifier that lines up with AMD’s future graphics roadmap. These documentation changes include initial support for GFX13, and notably, the entry explicitly associates GFX13 with the RDNA 5 architecture. While branding decisions can always change, this is one of the clearer signs yet that AMD plans to continue the RDNA naming convention for its upcoming Radeon GPUs.
So what is GFX13, and why does it matter? In AMD’s ecosystem, these “GFX” IDs are tied to specific GPU generations. With GFX12 currently covering the RDNA 4 era, the move to GFX13 represents a major jump—exactly the kind you’d expect when a new architecture is on the horizon. The leak also references a GFX1310 branch, which is believed to point toward AMD’s discrete desktop GPUs rather than integrated graphics. If AMD sticks with its current Radeon naming pattern, that would likely place GFX1310 in the territory of a future Radeon RX 10000 series lineup.
It’s important to keep expectations grounded: seeing early support land in LLVM doesn’t mean a launch is imminent, and these kinds of entries can appear well before products are publicly discussed. Still, they’re meaningful because they represent the behind-the-scenes groundwork required for drivers and developer tools to properly recognize and optimize for new hardware.
LLVM plays a central role in AMD’s Linux graphics and compute pipeline, underpinning key parts of the toolchain used across Mesa, ROCm, and other development and driver-related software. When a new GPU architecture starts appearing in LLVM, it’s a sign that the architecture is moving closer to being fully supported across the software stack—an essential step for performance, stability, and compatibility when the GPUs eventually launch.
As for when RDNA 5 GPUs might arrive, current expectations suggest they won’t show up before 2027. The architecture is rumored to use TSMC’s N3P process node, with a possible launch window around mid-2027. That timing could set up a direct collision with NVIDIA’s next wave of GPUs expected around the same period.
Details about the RDNA 5 product stack remain scarce for now, but many enthusiasts are watching for one key outcome: whether AMD will push back into the high-end GPU segment in a bigger way than it did with RDNA 4. If RDNA 5 represents a more aggressive return to flagship-class competition, the groundwork showing up today in LLVM could be the earliest hint of a much larger strategy taking shape.
