Apple are NOT trying to move away from Samsung. They moved to TSMC because they had 20nm available sooner, but will be moving to Samsung for 14nm (20nm + finfets) because it will be available 3-6 months before TSMC's comparable finfet process is available. Though it's said that TSMC's might be the best overall process, but specifically stated that it's factoring in cost. It's quite possible that Samsung/GloFo will be first to decent quality finfet on 20nm and have the best process, but cost a bit more.
Second sourcing is a HUGE deal in these industries, both for overall capacity and having separate locations. If there is an earthquake or a flood around a TSMC fab, with most of them localised it can stop output for months, think the floods causing HDD issues. For someone like Apple that could be billions lost in a quarter or two if no chips are being supplied.
One of the biggest reasons Samsung/GloFo hooked up to use the same 14nm process was precisely so Apple could get a second source, and Qually care about second sourcing to. They both moved/produced more at TSMC this round but moved back basically the second Samsung/GLoFo situation was sorted.
In terms of what gpu's will come out where. There is a general trend of AMD trying to use GloFo, but new processes aren't easy. Waiting 6-12 months to launch a card can make it almost obsolete.
While AMD's plan is to move more to GloFo, they'll do whatever works at the time. if GloFo's process sucks or is a little late for one process they'll just tape out at TSMC for that card. If GloFo are ahead for the next process step they'll switch right back.
GloFo's best 28nm process is incredibly close to Intel's 22nm in quality, and likely not too far off a low yielding new 20nm TSMC process, less dense, sure, but power wise maybe pretty close.
If 20nm TSMC, looking to be more like late 2015. If TSMC 20nm + finfets(marketed as 16nm), probably mid 2016. if GloFo 20nm + finfets(marketed as 14nm) then maybe late 2015.