According to the well-known source on NVIDIA graphics cards, kopite7kimi both NVIDIA and AMD are working on their next-gen graphics cards in earnest. The next generation of gaming graphics cards is expected to come with an expected performance gain of 2.2-2.5x compared to the RTX 30 series lineup. His sources have informed him that both the lineups, Ada Lovelace (NVIDIA) and RDNA 3 (AMD) will leverage a 5nm node, the former from Samsung and the latter from TSMC. Furthermore, he’s optimistic that we might even see a 100 TFLOPs GPU (FP32/SP) amongst the top-end parts. In comparison, the GeForce RTX 3090 is capable of “just” 35.58 TFLOPS of FP32 (single-precision) compute.
According to Kimi’s estimates, the top-end RTX 40 series GPU (Ada) will be more than twice as faster than the GeForce RTX 3080/3090, while AMD’s chiplet-based Navi 31 flagship (with possibly 2 compute dies) may be even faster, being up to 2.5x faster than the top-end Ampere parts. Finally, NVIDIA’s first MCM design (codenamed Hopper) is targeted to be a whopping 3x faster than Ampere.
Considering how far we’re from the launch of these next-gen GPU families, it’s important to remember that nothing is final, especially on NVIDIA’s end. If AMD manages to beat Team Green by a notable margin, as we’ve seen in the past, Jensen might as well work on another refresh or die for the enthusiast market.
