Based on wccftech and well-known {hardware} leaker @KittyYYuko, AMD’s next-generation Radeon graphics playing cards needs to be a large improve—as a lot as thrice quicker than AMD’s present flagship graphics playing cards.
2.5x is simply too little.
Take a guess :p— Yuko Yoshida (@KittyYYuko) May 2, 2021
On this reply to RedGamingTech’s Paul Eccleston, @KittyYYuko claims a 3x efficiency enhance when upgrading from AMD’s flagship Navi 21 (as present in current-gen RX 6800XT and RX 6900 XT playing cards) to Navi 31.
RDNA3—the structure beneath the brand new GPUs—will shift process-node dimension from 7nm right down to 5nm, which tends to lead to main efficiency beneficial properties and even bigger performance-per-watt beneficial properties. The downshift in course of node is not sufficient to account for the claimed efficiency uplift by itself, although—that comes from making the GPU itself considerably larger and meaner.
The present flagship, Navi 21, affords 80 compute models and 5,120 shading models. Navi 31 will reportedly double that, providing a whopping 160 CUs and over 10,000 shading models, delivered in two 80 CU chiplets on a single die. That is probably solely sensible because of the downshift in course of dimension—cooling and energy provide are main considerations in flagship GPUs, which are inclined to require as a lot of each as OEMs and particular person builders can comfortably cram right into a case.
RedGamingTech means that Navi 31’s ray-tracing efficiency needs to be roughly on par with Nvidia’s Ampere GPUs, however machine-learning duties will nonetheless be dealt with by way of low-precision operations on the primary GPU itself, fairly than on devoted {hardware} like Nvidia’s Tensor Cores.
We anticipate RDNA 3 playing cards to grow to be out there someday in late 2022 or early 2023.
