Quote From Eurogamer:
Storage is one area where Sony has a clear advantage over Microsoft. The PlayStation 5’s custom 825 GB SSD has a theoretical transfer rate of over 8 GB/s, factoring in its compression algorithm. This allows it fill up the PS5’s 16 GB of shared memory in just 2 seconds. The Xbox Series X on the other hand features an NVMe SSD that offers a guaranteed 2.4 GB/S of throughput. It’s important here to remember that neither of these figures is exactly lacking. 7200 RPM hard drives offer peak transfer rates of around 200 MB/s, which is less than 10 percent of what the Xbox Series X manages. 2.4 GB/S is blazing fast.
Nevertheless, the PlayStation 5 delivers over 2 times the nominal storage throughput and over 4 time the notional throughput once we factor in compression. The Xbox Series X’s SSD is evidently fast enough to enable features like Quick Resume and eliminating in-game loading screens, as well as the extended memory concept. At this point, we’re not quite sure how the PlayStation 5’s significantly faster SSD will or can be used, though, again, minimised loading times and enhanced asset streaming are likely to be key use cases.
In terms of memory, the PlayStation 5 offers developers a unified 16 GB pool of GDDR6, across a 256-bit memory bus for 448 GB/S of bandwidth. The Xbox Series X, in contrast, offers 10 GB of memory at 560 GB/S and a further 6 GB at 336 GB/S. While split memory architectures have had a bad rap (the GTX 970 and the PlayStation 3 come to mind), 10 GB is a lot of very high-speed memory. By using the 6 GB for less critical purposes, the Series X can utilise most of the high-speed memory for the GPU, and leverage that bandwidth to scale better at 4K. All in all, the memory configuration for the Series X is just better suited to the ninth-gen 4K resolution standard.
Basically, Sony's claimed performance relies COMPLETELY on efficiency of linear scaling in RDNA architectures, which RDNA1 was very poor at. Neither system is using 7nm EUV, so they won't enjoy the node benefits that process brings. However, that isn't to say they are using "raw" DUV the way RDNA1 does, either.
While the Eurogamer article suggests a real-world practical performance delta of upwards 30% in favor of XSX on the GPU side, it gives 20% on its low end, so it's probably safe to put the actual potential delta cap at around 25%, striking right in the middle. So the chances are real that the actual performance delta on GPU side is larger than the paper 15% - 17% we can work out through the spec numbers provided themselves.
But we shall see.
Another topic on this, we won't know anything until we get real world testing, but it is more like a 20-30% power difference as PS5 will not be able to maintain a boost clock speed under heavy load.
It's not so much if PS5 can maintain those clocks (I fully expect it to majority of the time; technically intensive games or poorly optimized titles might cause the 2% drop in clock frequencies Cerny mentioned), so much as it is the fact that simply clocking the GPU that high might not bring anywhere near the linear scaling in performance Sony would like to claim. We already have RDNA1 cards that show how poorly performance gains get once you start pushing super-high clocks.
PS5 will fare a bit better in that regard due to being on improved 7nm node, but it (nor XSX) are true 7nm EUV, and there is probably good reason why MS locked their GPU clock to 1825MHz; that is probably a little around the north spot of the sweetspot range for RDNA2 on their improved 7nm process. The northern end of the sweetspot may be higher on EUV, but again, neither console is using EUV.
So we have a situation where Sony is still pushing well past the sweetspot range, otherwise I venture MS would've felt comfortable clocking XSX's GPU even a tad higher. Although yes, they have a bigger chip, that plays into it as well, but again we can look at RDNA1 cards with chips the size of PS5's and see how little performance gains they got in real-world applications with much higher clocks.
It will mainly come down to how Sony's 1st party can work around the issues, but it's suffice to say the GPU performance delta between PS5 and XSX might be noticeably larger than 15% - 17% in real-world applications due to the factors I've just mentioned.