For me, it feels like Microsoft are confident in their position and can do this gen+ rasterization games at 8k30. And probably some indoor confined area ray-tracing at 4K30, along with a TF metric (eg 12TF RDNA) that is on par or better than Sony and will announce when Sony actually shows their hand.
With Sony, it feels more like we're currently playing a 'secret sauce' treasure hunt, where the hunt only ends when someone finds the secret sauce.
So here's my wildly unrealistic guess about Sony's secret sauce.
Sony have talked up the SSD from the start, because it is more than just hierarchical SSD storage for PS5, and is how they'll secret sauce the paradigm shift to RT without massive brute force.
I'm guessing that AMD got the direct GPU to SSD solution (for Radeon Pro SSG) from Sony and the design came about through Mark Cerney looking at the Voxel GI technique that was removed from Unreal at the start of this gen.
If Mark concluded that such GI techniques really needed a large virtual VRAM scratchpad, of say 64GB – an on GPU SDD - he would then be expecting to use highly compressed, precalculated space partition structures (for the RT) to be dynamically loaded from disc/mechanical hdd into the ssd and then be rapidly expanded into virtual VRAM with a small GPU overhead. The modest 9TF metric would then be plenty for high quality RT GI in real-time. And by choosing an SSD on GPU solution Sony would have stole a release window of graphics cards to define RT image quality in mainstream gaming, until SSD on GPU became the norm in PCs or a cheap workaround was found.
Obviously developers would have been kept in the dark about the SSD on GPU by H/w virtualizing the solution, and presenting it as a hierarchical storage feature to enhance clipmap performance ala Spiderman demo.
