Dunno how true of a case this actually this. First off, we know RDNA2 is a dual-process initiative: 7nm DUV enhanced, and 7nm EUV. We know the consoles are on 7nm DUV enhanced. We also know there have been persistent rumors for about a year now regarding PS5 cooling, both in terms of outright rumors (not getting into that "recent" one which has a questionable source), and just inference from looking at the design itself.
RDNA1's sweetspot is between 1700 MHz to 1800 MHz for power-to-performance. While AMD's mentioned 50% PPW gains in RDNA2 over RDNA1, we don't know on which process they are specifying this, and there's a chance there's some slight embellishment to that figure, because these kind of companies (AMD, Nvidia, Intel, Sony, Microsoft etc.) do that all the time.
If I were being generous, RDNA2's sweetspot probably has an upper limit of around 2 GHz, and that's with sufficient amounts of cooling being provided (comparable to an upper-end PC GPU card most likely). We can look at the XSX and see that even with "just" 1825 MHz GPU clock, that thing has a massive heat sink and a pretty elaborate cooling setup of its own. And, again, it's the same process node as PS5's GPU.
I'd expect PS5's cooling (in terms of sheer amount) to be around that level, because even if the GPU is smaller that's being offset by the higher maximum clocks. But I would still say that for an RDNA2 DUV enhanced chip, PS5's GPU clock is most likely definitely pushing beyond the sweetspot. What we need to know now, is if the frequency-to-performance gain ratio has improved significantly from RDNA1 tech, because those required a vast amount of additional power beyond their sweetspot for marginal frequency gains.
Going by Cerny's own words, a 10% reduction in power will result in a 2% drop in GPU and CPU frequency. So that's a 5:1 ratio, essentially, of power-to-frequency. I don't know how that compares to RDNA1 tech, but I'd picture that you'd want a smaller ratio than that. That amount he refers to could be referring to a clock range beyond the RDNA2 sweetspot, as well, because I would assume with overall architectural and node efficiency gains you wouldn't need a 5:1 ratio at sweetspot ranges (or certainly anything lower than that).
Oh well; we're going to see eventually once Sony does that cooling teardown (I'm assuming with Digital Foundry and maybe a few other tech channels as well). Ultimately this isn't me speaking to anything in the article, not agreeing or disagreeing with anything there. But I just wanted to touch on the power, heating, and cooling aspects of PS5 and RDNA2, just some personal speculation.
ethomaz
who brought up an interesting point regarding power-to-frequency scaling on RDNA1 and speculated possible efficiency gains in that regard with RDNA2. Their reply was the following:
