Not that particularly. More like, the model is good enough to run off compute units which don’t run in parallel to rendering, so if there is sufficient time for it, a different model could conceivably run on series X.
It is that way today for fsr3 for instance, except FSR3 doesn’t take any advantage of the DP4A of XSX. but we have a strong signal that PSSR could be using that and some (if running on CUs). The main advantages here by 5Pro would be the dual issue combined with sparse format support giving it 4x the capability.
I’m 50-50 on it, you can math out compute units to be able to generate 300TOPS for sparse format support into silicon, but XDNA may also be small enough to be integrated into the chip.
I really don’t know, I’m leaning towards the latter, but it looks like we have to wait to know the truth of this one.
TLDR: XSX could have a better model than FSR3, but not better than PSSR. But if PSSR requires XDNA, I don’t think there is any chance we will see custom scaler from MS.