It's logical to assume that splitting pure compute from memory controller & infinity cache would lead to increased clocks but how much more was difficult to assess because we didn't know how these 2 affecting each other.
GPUs always had multiply internal frequencies depending the segment, for example infinity cache is connected to the Graphics Engine at 1.94GHz (boost clock) in Navi 21.
The same leakers that were saying that Navi 21 hits 3GHz the same now are saying close to 4GHz (close is what 3.9GHz?) for Navi31, but even If this rumor is correct we shouldn't exclude instead of 2250MHz that RX 6900XT was, something like only 2925MHz boost even if near 4GHz is true.
On the other hand if Navi24 at 6nm could hit 2.815GHz turbo I don't see why at 5nm not to be able to go at 3.1GHz, now add the clock benefits from compute/memory controllers+infinity splitting, so even higher than that.
But all depends from TBP, at 350W the reference boost clock shouldn't be more than 3GHz anyway I would guess.
Still near 4GHz when pushed to the absolute limit is a surprise (if true...)
