At least on hard data that's been provided out there (photo evidence, expert analysis and GPU testing benchmarks), the only way XSX would hit below 10TF Navi on a 56CU core would be if the clocks are set below 1400MHz...which in that case, why even go for a tower design? A more traditional system design would be more than adequate. Either that, or they've disabled something like 12CUs which...why go for a die over 400mm2 in the first place?
It just wouldn't logically make sense in either case. Now, based on those same hard data points, it's possible PS5 could be around or shy of 10TF Navi; either they turn on more of the disabled CUs and be very careful with the high clocks, or they keep 36 active and push those clocks as high as humanly possible with some super-eloquent cooling. Both are risky and potentially unrealistic scenarios, but again, Ariel and Oberon are the only PS5-related chips we've seen consistent benchmarking tests and information on. If there's another PS5 GPU chip around, I would sure love at least some proof of its existence in even a partial benchmark; heck we've even gotten more proof of possible Lockhart GPU testing than this hypothetical PS5 GPU chip.
I just like going with logic and observable data, especially if there's a pattern to that data. Which is why I'm more willing to entertain the Oberon/Ariel benchmark tests than rumors that, interesting as they are, don't have any substantiated data backing them, especially this late into the process for the new consoles. But maybe something of that effect will come around between now and whenever the PS5 event is, in which case I'm gonna look forward to that (well, I'm looking forward to the even regardless, but you get the point).