According to wccftech and well-known hardware leaker @KittyYYuko, AMD’s next-generation Radeon graphics cards should be a massive upgrade—as much as three times faster than AMD’s current flagship graphics cards.
2.5x is too little.
Take a guess :p
— Yuko Yoshida (@KittyYYuko) May 2, 2021
In the above reply to RedGamingTech’s Paul Eccleston, @KittyYYuko claims a 3x performance increase when upgrading from AMD’s flagship Navi 21 (as found in current-gen RX 6800XT and RX 6900 XT cards) to Navi 31.
RDNA3—the architecture beneath the new GPUs—will shift process-node size from 7 nm down to 5 nm, which tends to result in major performance gains and even larger performance-per-watt gains. The downshift in process node isn’t enough to account for the claimed performance uplift by itself, though—that comes from making the GPU itself significantly bigger and meaner.
The current flagship, Navi 21, offers 80 compute units and 5,120 shading units. Navi 31 will reportedly double those numbers, offering a whopping 160 CUs and over 10,000 shading units, delivered in two 80 CU chiplets on a single die. This is likely only practical due to the downshift in process size—cooling and power supply are major concerns in flagship GPUs, which tend to require as much of both as OEMs and individual builders can comfortably cram into a case.
RedGamingTech suggests that Navi 31’s ray-tracing performance should be roughly on par with Nvidia’s Ampere GPUs, but machine-learning tasks will still be handled via low-precision operations on the main GPU itself rather than on dedicated hardware like Nvidia’s Tensor Cores.
We expect RDNA 3 cards to become available sometime in late 2022 or early 2023.