It's hard to extrapolate because the gaming core is different, so we don't know what changes have been made to tensor or RT cores, any cache changes there and clocks. I can accept Cuda cores maybe have the same IPC as Ampere but you still can't use that to estimate performance because you need more info than just IPC of the Cuda core
that said I'm not surprised, very early on we had rumours that Lovelace would just be a bigger Ampere.
Lovelace I guess will be 5nm, not 4nm unless Nvidia got a really good deal
Fair point. Just a rough extrapolation on the CUDA performance rather than gaming performance.
TSMC 4N is a revision of the N5 as well (N7, N5, N4, N3 are TSMC generations, whereas 7N, 4N are rebranded TSMC nodes for Nvidia), not a big difference between them. There's a huge demand for TSMC's 5-4nm wafers between Apple, AMD, Qualcomm, Mediatek and AMD this year, so it's not going to be cheap. I can see Nvidia going for Samsung again.