There's a chance this patent might not be for PS5 at all, but the PS4 CPU. However, if it is referring to unified L3$ on the CPU then I can see that being the case. It makes a lot more sense than some phantom Infinity Cache in either system. For heavy CPU-bound tasks it could be of some benefit to PS5 relative to Series X, but I wouldn't expect night-and-day differences in that regard.
As to if this is to explain for some of the performance differences in 3P games between Sony and MS systems so far...well the mistake people make too often is looking for a "smoking gun". There is no "smoking gun", it's not that simple. I'd say unified L3$ is an exacerbation of the issue in that regard, not the root of it. Same would've been the case if there was IC on PS5 (there isn't). Even something like one being variable clock and the other fixed clock, in the end those are just exacerbating the root cause.
And what is that root cause? Combination of new API elements in the GDK, some GDK features still being unstable, certain functionality like NGG not fully going yet, and lack of time in familiarity with GDK features by some 3P devs, as well as perhaps lack of time to optimize the Series versions (PS5 versions may've been the priority), lack of timely GDK for Series S potentially impacting Series X development/optimization.
The total net of factors are a lot more complicated than just pinning it on one thing, especially one thing at a hard-fixed silicon level which would immediately imply the person going right to that conclusion doesn't think (or doesn't wish) the current ailments can be fixed with effort and time. Generally you need a lot more time and results to even begin pinning certain performance issues to very specific hardware limitations anyway, and even if such limitations turn out to be the case, there are still devs (usually 1P) who can work around those and produce results commonly accepted as impossible with those limitations present, through smart design and programming.
That 4A dev article is from 2014 - a year before DX12 was released. DX12U introduces even more changes in terms of draw calls. I don’t think this take is super relevant, especially once the Series X tools mature.
MS consoles also have extended hardware features for executeIndirect explicitly for handling drawcalls, this would be present in the Series systems as well.
It's kind of like people clinging onto super-old John Carmack quotes from the mid-2000s on narrow/fast vs. wide/slow designs, not understanding those quotes were made in a specific era and within the constraints of technology of that era. They aren't necessarily applicable as blanket statements in today's technological environment. Nuance in parsing the meaning where appropriate has to be taken.
Amd used unified L3 in the past in their processors without sucess
L3 unified Cache doesn´t mean that it´s better. Its mosly depending of the general desing
Yep. When people hear something they think sounds new and therefore exotic in some way (I'm talking tech-wise), and they may already have some common-formed perceptions about two or so pieces of kit relative to each other (assuming one must be pedestrian, and the other must be unique), depending on their various preferences brand/technology-wise if that particular "thing" falls on one side or the other they will attach either positive or negative connotations to it.
But this usually is done without actually researching the feature in particular from a wider industrial field perspective to see in what ways it's actually been implemented and in what ways it's been useful (or a hindrance). Now with all that said, something that maybe was implemented in the past and didn't work out too well, doesn't mean the idea or concept itself is fundamentally flawed. Maybe its time for practical use was just too early when first utilized. So maybe the bigger reason why unified L3$ on AMD's older CPUs didn't bring much wasn't because of the unified L3$, but because AMD's CPU microarchitecture wasn't optimized for leveraging a unified L3$ efficiently back in that time.
It could, however, now be optimized for that with Zen 3, so they bring it back. Stuff like that.