You forgot the most important parameter. If the rumors for 4 HBM chips that will give a 4096bit data bus and 640GB/sec bandwidth are true, then both gpus and APUs might see a good performance improvement before 2016 when Zen will comes.
As I said on another thread, HBM gives a wide I/O and high bandwidth. HBM is also limited to 4GB (4 x 1GB stacks), so there are a few things to consider:
1. Wider I/O means more contacts and traces. Higher cost (so you're only looking at the flagship boards anyway).
2. 4GB framebuffer isn't a good marketing point when even mainstream cards will likely sport 4GB as standard, and 8GB cards at the high end are being introduced. IIRC, the GM200 articles allude to 12GB vRAM. Assuming AMD see it the same way, they may have to adopt a tiered memory structure (HBM + GDDR5) which will be more complex (differing latencies, circuitry/traces), more expensive, and wipe out any power saving claimed by HBM alone.
Just how popular are high end cards anyway, if GTX 9xx is behind this? AFAIK most action (fiscally speaking) happens in the low-mid range, and there AMD are excellent.
Doesn't really matter at this stage. The game is more about mind share than market share. Once upon a time, those of us around at the height of ATi's powers saw a company that footed it with the best - at one stage the company had the lions share of OEM contracts and annihilated S3, Matrox, 3dfx, and Nvidia in sales revenue. It was/is a hardware market driven by features and marketing because it markets directly to the end user. Processors on the other hand began as companies selling to the engineers of other companies - marketing was something you did when the product couldn't compete. AMD was founded by a salesman (and a group of analog logic circuit designers IIRC) when marketing wasn't really a thing in CPUs - by the time it was, Sanders had moved on and engineers held the reins at the company. It isn't a complete coincidence that as soon as AMD took over ATi the graphics side started to dip. All the work the company had put into a strong relationship with the gaming community went for nought - starting and stopping the Get in the Game program just as TWIMTBP was gaining traction for example. When your star is being eclipsed (by Nvidia), the response should have been a full court press, but AMD withdrew and began revelling in it's "underdog" status after R600. When you lose the customer base the OEM's soon follow since they pander to what is a saleable and easily recognizable series of brands, and without the OEMs onside a company becomes severely handicapped. The DIY discrete market is miniscule, but the OEMs control virtually all PC desktop sales, and all mobile sales. Then it becomes a vicious circle, OEMs highlight the brand of the company in ascendance both in advertising and model lines and the other companies are marginalized, thus slipping out of the consumers field of view.
Why people still count Intel into the bunch. They count every single GPU component as market share even though tons of their chips never ever display a single frame of image on any screen. But they always come with their CPU's whether you like it or not.
AMD are in the same basket. MANY systems (especially mobile) rely upon integrated graphics. If the iGP is competent enough to do the job (and a lot of people either don't game or play garbage flash based "games") then they have no need of a discrete solution.
Count AMD's blessings. I'm pretty sure when Mercury and JPR release the discrete graphics numbers next week the figures will look even more dismal given that it's a two horse race for a smaller market. It's also pretty safe to assume that the ASP (average selling prices) of the two companies are heading in opposite directions coming into the busiest sales period of the year.
The original chart in the JPR press release had the flawed numbers. I checked a few sites and JPR itself just after the article was published and all the sites carried the same borked chart numbers - which was, along with the wording, repeated verbatim from JPR's PR.
