Following talk of Nvidia GeForce RTX 40-series GPU architecture being tailored depending on how AMD's RDNA3 graphics cards looked, further rumours have surfaced to suggest the project planning stage has now been finalised. There are even speculations concerning a rough launch date for the next gen Nvidia graphics cards.
Posts scattered across Kopite7kimi (opens in new tab)'s Twitter account (a renowned leaker), already had Nvidia's GeForce RTX 40-series cards pinned as using TSMC's 5nm process nodes. This came along with hints that the new cards could end up switching from AD102 Lovelace architecture to GH202 Hopper architecture, depending on how AMD's RDNA3 plans unfolded.
Now though, another known leaker Greymon55 has chimed in, convinced that the 40-series project planning stage is over, and that the design stage will soon be underway. If true, that would mean the (Ada) Lovelace architecture could now be locked in.
Ada TSMC 5nm 100%, but I'm not sure if it's N5 or N5P yet.July 24, 2021
In reply to questions about the expected process nodes, Greymon55 is adamant that we'll be looking at TSMC's 5nm, and Lovelace architecture, though they remain unsure whether this will be standard, or the optimised N5 Performance-enhanced version (N5P). Greymon55 also speculates a release date (opens in new tab): "No earlier than the end of 2022."
Further rumours (via TechRadar (opens in new tab)) place the 40-series flagship GPU at 144 streaming multiprocessors, and a maximum of 18,432 CUDA cores, should it be based on the AD102 architecture. That's almost double the RTX 3090 (opens in new tab)'s 10,496, and though the different architectures aren't directly comparable, those 7,936 extra cores would make for one hell of a beefy chip.
Either way, Nvidia's 40-series cards appear to be sticking with a monolithic design, as opposed to following AMD's multiple IC MCM architecture approach.
Just remember to take these early rumours with a grain of salt. Nothing is confirmed just yet from the company itself, so we recommend waiting for Nvidia to make an announcement, rather than taking any leaker's word for it at such an early stage.