Each CU has 64 Stream Processors (SPs), for a grand total of 4,608 SPs on the Radeon RX 6800 XT. Other than the new Ray Accelerators, each Compute Unit (CU) is laid out much like the original RDNA cards.
We'll dive into just how effective these are later on, but we expect performance to improve as AMD's drivers mature – much like the improvement that Nvidia's RTX cards saw throughout the lifespan of the Turing cards. There are one of these on each of the 72 Compute Units, which is what lets the AMD Radeon RX 6800 XT be able to handle real-time ray tracing at a playable framerate. The other major addition to the architecture is, of course, the DirectX 12 Ultimate-compatible Ray Accelerators. Outputs: 1 x HDMI 2.1, 2 x DisplayPort 1.4a, 1 x USB Type-C This helps narrow the gap that AMD would otherwise experience when going against the GDDR6X memory in flagship-class Nvidia cards. This helps feed data from the larger 16GB VRAM pool to the GPU, and this combination of speedy cache and GDDR6 memory results in 2.4x more bandwidth per watt than the VRAM would be able to achieve on its own. The Infinity Cache is basically 128MB of global cache, which AMD asserts is 3.25x faster than GDDR6 VRAM on a 256-bit bus. The biggest of these is the new Infinity Cache, which is crucial when you consider that the Radeon RX 6800 XT is using 16GB of GDDR6 memory, rather than the GDDR6X memory found on the Nvidia GeForce RTX 3080 and RTX 3090. Rather than shrinking the die, AMD has worked in some new features and boosted efficiency, which makes these graphics cards way more powerful than their predecessors. The AMD Radeon RX 6800 XT is based on Team Red's RDNA 2 graphics architecture, which is still based on a 7nm (nanometer) manufacturing process.