The FX CPUs were very forward thinking, designed for software that simply didn't exist at the time and ignoring the fact that such software would take some time to be developed. Multi-threading is very hard to do right, I know, I do it professionally, I can spend days debugging the most innocuous looking code. Strangely AMD made the same mistake with Hawaii and Fiji, a hardware that in theory should be incredibly powerful but hammered by the current APIs and some frontend bottlenecks. This is something of a theme for AMD - tomorrow's technology today. Sounds great, forward thinking, but in reality always comes back to bite them. HBM, great stuff, but utterly pointless on a FuryX, Pioneering Tessellation, and the totally ignoring it when it becomes practical.
Intel made a similar error with the Itanium. X86 architecture sucks and is far to antiquated, Buick on crumbling bloated foundations. Itanium-64 is fantastic improvement in theory. In practice they never made decent compilers, making decent compilers was incredibly tough, and even when they did it was a failure because no software developers would code specifically for that architecture.
Ironically AMD by choosing the cheaper and simpler 64bit extension x86 was a fantastically successful strategic move that really kicked Intel down a beat.
yes and no, it was forward thinking, but if you remember few year prior, there was a lot of talk about DX and multi-threading, but DX10-11 came out not very far from DX9, microsoft didnt follow, and didnt seem to have plans to change that anytime soon, which prompted the birth of Mantle, but that was like 4 years late, enough time for AMD CPUs to get the reputation of being bad, which they are not if put in their element.
about GPUs to me thats not a mistake, they did it when they decided to do custom soc specific for consoles, and it paid out, so they had to sacrifice a little bit on the PC front in efficiency for that, but the upside is that they are in every console on the market, Devs are more familar with their hardware, they are in the middle of it being able to influence change from the inside, and their next step is pushing multi-gpu into console, and laveraging that on PC by making multiple chips on an interposer.
now that the API is less crappy than before, they are doing better with software using them on both fronts gpu and cpu.
to be honest AMD did really well, with so little money, and really bad luck during these last years, most companies would have gone belly up long ago, or gave up, they have a really good long term planning, so polaris/vega/zen should help them get out of the red zone definitely by the end of the year, at E3 if sony and microsoft announce their new consoles upgrade and turns out to be multi-gpu, then AMD is going to have a great 2017/2018, from market share to margins, and money will flow...hopefully