Looks like this card is going to be even better value once developers add proper Vulcan support:

https://www.youtube.com/watch?v=WOaHpZjQ73M

RX 480 now easily beats a GTX 980 in Doom and the Fury X beats a GTX 1070, even the R9 Nano is faster at 1080P.

Saw that earlier. Nice rise from the dead for the Fury X, which has always been sold above GTX1070 prices so its great to see it finally pulling its weight for its cost.

The question is will there be a better example where this matters than Doom? Getting 60 fps in Doom at quality settings is not quite an issue for many cards.

I would like to know how many upcoming games are set to be developed from the ground-up on Vulkan, besides idTech games.

Only to put things in context, AMD has released a new driver optimization for Doom and Vulkan while Nvidia still hasn't done it. That means that, right now, AMD cards have advantage.

About process, yes, I know, and I was meaning that maybe the whole range could still receive tweakings. We know that 480's power problem was of power management, taking too much power from PCIe instead of from auxiliary power connector, but also after fixing that, we see that power efficiency is quite lower than promised, and while it's quite normal that real world performance be lower than ideal ones, maybe we can still expect improvements not only from drivers and from improved board design, but also from the HW itself if the production process can still be tweaked.

About APUs, I wasn't precise, I should have written the use of the latest AMD GPU cores in them, and not literally putting a 460 in a Zen APU, but simply putting the GPU power of a 460, obtained with the latest cores available, in it. Yes, a sideport for GDDR could be a nice idea to avoid wasting the GPU power of the  higher-end APUs, although faster DDR4, and organised in quad- or even better eight-channel architecture, could at least mitigate the problem.

About drivers and games, thanks for all the explanations, my doubts come from the existence of games that cannot use more than one GPU, or cannot without giving more problems than benefits, maybe it could be possible to transparently offer them at least part of all the benefits offered by multiple GPU with APIs that keep where it should be stuff that is at a level low enough to be better managed by those that designed the HW than by game devs.

The entire 300 series was rebadges though with Fury added on top as a "Halo" product and a test vehicle for mass produced HBM and Nano reserved for a more Niche' market.

The 390 and 390X were just re-badged Radeon R9 290 and 290X parts with the bulk of the 200 series being rebadged 7000 parts.

AMD has just stagnated, not just in terms of rebadged hardware, but also prices and that reflected in their marketshare, they are turning that around... But it will sadly take another couple of years to see the fruits of AMD's new strategy.
Even the 400 series will likely be derived by a substantual amount of rebadged hardware, especially in the low-end.

If Vega doesn't launch untill next year, it might not launch under the 400 series lineup, but rather the 500 series instead, with Polaris and general GCN 1.0/1.1/1.2 rebadges.

AMD *could* also launch two dual-GPU cards.
With the Radeon 3000 series and 4000 series AMD actually released multiple x2 cards one based on their fastest GPU and the other on a slower GPU. Aka. Radeon 3850 X2 and 3870 X2 and 4850 X2 and 4870 X2. - Could be a return to form and we end up with both.

Well, it goes without saying that there will be some refinement in production as time goes on that should result in some reduction in power consumption after a few revisions.
But it's not going to be anything significant...

For something more significant AMD will likely need to do another respin and that won't happen for a long time yet, if ever.
Basically the best refinement we can except is from custom cards.

Quad-Channel and Octo-Channel carry with it increased costs due to a substantual jump in PCB traces required, which means more PCB layers and thus more engineering to route everything properly.
It's not gonna' happen. :P Not in a consumer-grade APU anyway.

DDR4 and the bandwidth saving technology AMD has implemented in GCN 1.3/4.0 will help, but it's not going to allow you to have high-end GPU performance, mid-range maybe. Low-end certainly.

That is where Microsoft is taking Direct X 12. Game developers can build their games to support their own multi-GPU implementation, this is what nVidia has backed when it comes to SLI support for more than 2 cards in games.

But... Leaving it in the hands of developers means it's never going to catch on considering that the vast majority of PC's use a single GPU and consoles all use a Single GPU, it's a waste of resources.

At the moment though if you want more than 2 GPU's for gaming, AMD is where it is at.

There's no need to go that back, AMD launched the 295X2 with two Hawaii XT chips and this year they finally launched the Radeon Pro Duo with two Fiji XT processors.

A dual Polaris card won't surprise anyone, it's only the name that's open to discussion. RX 480X2? RX 480 Duo?

3Dmark's Time Spy DX12 Dem is out today, and Rx480 on that API get near a GTX980 without Asynchronic Computes, and Above that high end GPU with it activated. Really nice for a 240$ GPU. here some tests:

http://www.pcworld.com/article/3095301/components-graphics/3dmark-time-spy-tested-we-pit-radeon-vs-geforce-in-this-major-new-dx12-benchmark.html

I wouldn't put too much faith in that Tweaktown article, even more so containing bits like this one: "We know from AMD's revised naming system for the Radeon series that that the RX 490 will feature a 256-bit memory bus and will be aimed at 4K and VR gaming."

Looking at AMD's new naming slide, any 490 card will have a higher than 256bit bus memory, not a 256bit bus one. They are clearly writing their thoughts/guesses as facts.

Now, on another note:

SK Hynix to Ship HBM2 Memory by Q3-2016

https://www.techpowerup.com/224149/sk-hynix-to-ship-hbm2-memory-by-q3-2016

Korean memory and NAND flash giant SK Hynix announced that it will have HBM2 memory ready for order within Q3-2016 (July-September). The company will ship 4 gigabyte HBM2 stacks in the 4 Hi-stack (4-die stack) form-factor, in two speeds - 2.00 Gbps (256 GB/s per stack), bearing model number H5VR32ESM4H-20C; and 1.60 Gbps (204 GB/s per stack), bearing model number H5VR32ESM4H-12C. With four such stacks, graphics cards over a 4096-bit HBM2 interface, graphics cards with 16 GB of total memory can be built.

*Wets Pants*

Vega here we come!

yea, the Maxwell gen cards will lose a lot of ground against competing AMD cards in newer APIs as they were extremely finetuned for DX11 performance