‘AMD Navi chips based on RDNA 2 have 40 and 80 compute units’ – Computer – News

No, that is a very strange leap you make there in your argumentation.

No, that is certainly not a strange jump, that is just the logical consequence. With better design, AMD can clearly make a much more efficient GPU at 7nm. That potential is simply there. That is literally undeniable because nvidia has already shown that it is possible.

But for some reason you immediately assume that RDNA again has an inherent bottleneck like GCN, when RDNA is designed to get rid of it. That is a really strange jump.

and that AMD apparently has almost not exploited that stretch with RDNA 1

Since they literally only made 2 chips out of it, and that they were very busy with the console GPUs at the time, and they still have limited resources, that would really not surprise me at all. RDNA was good enough to keep track of nvidia in perf per transistor, and be even in perf / watt, and could be a lot cheaper than nvidia’s nearly twice the chips, so that was more than enough for an intermediate phase.

(even though you previously claimed that only 15% of the 50% of the die shrink came and so a step of 35% had already been taken purely on architecture).

it was 25% (I had edited later, I also added the slide if you look back)
but 25% of the total.
So 12.5% ​​of the 50% if you reason it that way.

and the only substantiation you have is a marketing slide from AMD?

A marketing slide with a claim in the same vein that they have made more than true. and again, a claim that is not unreasonable at all.

If the 5700xt has the same perf / watt as nvidia at 16nm you could say that AMD 2 nodes at perf / watt was behind nvidia because of design.
If nvidia can make such economical designs then such designs are simply possible and realistic. And then AMD can do that too, especially because they had the R&D dollars available from both Sony and Microsoft for RDNA2, while RDNA1 was clearly a relatively short-lived intermediate release.

–