Honestly, it's hard to make heads or tails of this. It's probably not a refresh of P11 because they also confirm a refresh of P10 that's named differently
This is why P12 is likely a new chip. If their revision to P10 is called P10 XT2, then why would a revision to P11 be called P12?
I don't think it's something smaller than P11. Is it worth R&D and manufacture of something smaller than the cutdown P11? Who on earth is the customer?
In-between P10-P11 is possible, but I'm not sure what the point is. A fullsize, smaller chip (say if P12 is 1536SPs) would compete better with GP107 (larger P10 costs a lot more, slower P11 is just that - slower), but is it really worth it for the $120-$160 segment? Really?
The linked in leak confirmed (using that word lightly) 4096SP Vega chip. We still do not know if that's big or small Vega. If it is small, then there is plenty of room in between P10 and Small Vega for Polaris 12.
2304SP 32 ROP 256-bit. 4096SP 64 ROP (presumably) HBM2, and a
3072SP 64 ROP 384-bit Polaris 12?
That is simply my guess. 2816SP Hawaii and 2048SP Tahiti/Tonga coexisted. 3072/2304 is a very similar ratio.
Scorpio GPU could be a sign of that, but not necessarily. IIRC, PS4 uses a 1280SP (cut down to 1152) custom chip that has 8 ACE's, essentially making it a GCN 2 Pitcairn. But AMD instead milked the immortal Pitcairn and never released something like this.
But... they may be more likely to create a similar dGPU this time around because there's nothing old to milk if there's a big gap between P10 and small Vega since this is a new node. Not gonna keep making Fiji, obviously, and that's the only AMD chip that will fit between P10 and small Vega performance (if small Vega is 4096SPs).