My understanding, is that, in fact, AMD's GPU development (for their PC GPUs), was largely funded / provided by their R&D for the consoles. Clearly, even if MS and Sony are each paying for development, AMD retails rights to sell GPUs into the PC market with that tech in it, presumably co-developed. (Edit: Not speaking as if this is a proven fact, this is what I have read and have understood. I don't know how factual it is.)
I think your understanding is not correct and is actually basically the inverse of how it actually happens. AMD is already developing these GPUs, and then Sony and Microsoft pay AMD to tweak it and then they get exclusive rights to those versions of the chips for their consoles. AMD has consistently talked about the customization they've done for Sony and Microsoft. Microsoft talked quite a lot about about their own IP that went into the Scorpio chip (and if I remember right, specifically said it was different from Vega or the PS4 Pro), and there were parts even in the GPU that Microsoft owns the IP rights to (not sure how much, but they were talking about some DX12 draw call processor thing, but I believe there's more than that). Which that doesn't mean they would keep AMD from making a PC chip (Microsoft would arguably benefit either way), I just don't see it happening.
And it will muddy things up too much as both the PS4 Pro and Scorpio chips have substantive differences from Polaris and Vega, which would require special driver tweaks to take advantage of and/or possibly bottleneck (thus making them perform worse than in the consoles). That's not a smart move for AMD, where the software side of things is probably the biggest area they're lacking compared to Nvidia. On top of that, they're in the process of moving to a new architecture in Navi (its supposedly the first post GCN architecture, seeming to indicate there's pretty significant changes to it). Navi should be getting their resources, and they should be aiming to make it as good as possible and setting up their future GPU stack.
I actually wish AMD had lobbied Microsoft to offer the One X as a standalone SFF Win10 PC (really hope they do that for the next Xbox where the stronger CPU would make it even more feasible).
I agree with this, but I still see a pure cost justification for moving updated Polaris or Vega chips (basically, existing, quick-to-move tech) in the meantime if, for no other reason, it's actually the lowest-cost option for meeting the current WSA with GloFo.
Even if these don't sell all that well, it is probably cheaper for AMD to just buy a bunch of wafers from Glo Fo now, for chips that will, at least, sell, than worry about paying them off a year from now when moving brand new stuff with TSMC. No one (afaik) knows what the re-worked WSA will look like, so I maintain that as we are all pretty ignorant about these details right now, pushing out these low-rent, low cost, whatever chips could very well be tied to meeting the terms of that agreement on a purely overall cost basis for AMD, as they probably want to be done with Glo Fo ASAP for their bleeding edge stuff.
AMD earnings are tomorrow and I suspect that terms of the newest renegotiated WSA will be made public.
I don't agree. AMD has limits to their resources, so the more they spend doing stuff like that, the less they'll be spending on other things. Plus, they can do that without putting in the resources to do all this other stuff though. I still think this 12nm Polaris chip is mostly about supplying OEMs, and that they simply had to go to 12nm as GF likely has migrated their 14nm process to 12nm (since my understanding is, its entirely superior, in performance but also cost for GF as its really just 14nm tweaked, which included lower production costs for GF).
AMD's best move for the wafer agreement would be to give OEMs tons of Ryzen 2000 series and these new Polaris chips, for cheap. If they were to offer a deal where they go if you pair them together, we'll give them to you for almost cost (this way it makes sure that people get solid systems, and they have tiers in already, RX 660 with 4 cores, RX 670 with 6 cores, and RX 680 with 8 cores). That will move the most chips and bring the biggest benefits. It would be a big boost for budget consumers and could win them mindshare and should definitely help with marketshare. Plus it would win them favor with OEMs which is important and AMD has cited OEM deals as providing financial stability. It will probably be mid-year before most OEMs have Ryzen 2 and Navi systems, and those will carry some premium, so they'll still be able to sell plenty of cheap Ryzen 2000 (Zen+) and Polaris 12nm systems.
Moving forward, I think AMD should be looking at a different way of handling the WSA, and target FDX stuff. Be that for chipset related stuff, maybe partner with a company making communication chips to make something on GF's FDX (so AMD leverage their engineering and know how of GF's stuff, while companies on older stuff get to migrate to newer process easier helping them), and then integrate it into AMD systems (looking ahead, 5G modems on 12FDX for instance; in the meantime a wifi/Bluetooth/GbE for 22FDX). Another one to target would be a wireless chip for VR/AR headsets and/or displays (which that would also be a good use for Freesync, where it can sync frames based on the latency of the wireless or something so it can smooth video that would be really choppy otherwise).
I'm betting on the "stopgap counter" theory. Nvidia has the money, and the midrange market is still important, so I will bet they did this to act as a spoiler for the RX 590.
This also lends more credence to the theory that the RX 590 is a real product. Nvidia wouldn't do this just for the fun of it.
I'd guess that the GDDR5X cards are actually 1070s (and that its going to be going for the market that the 1060 originally did, maybe they drop clocks to up the efficiency or something while leaving OC room for AIB cards. This way it is a bit more than a simple rebadge, but they can sell a 1070 with GDDR5X for $300. The 1080 could hit $400 MSRP. The 1060 goes to $200. Maybe these become the 1100 series (so that it delineates between the RTX stuff even more, but marks a clear change over the 1000 series).
I personally have a hunch that we don't get RTX lower tier cards (other than possibly a cut 2070) until 7nm. It adds too much cost, and there's been a fair amount of backlash and the touted features aren't exactly offering much so far. I don't know if they could do simple ports of Pascal to 12nm, maybe add GDDR6 memory controllers, but I believe they have a significant stock of Pascal already in the channel due to them overproducing massively for crypto mining, so the best way to move them would be to drop prices and rebadge.
