So if they try to make their GPU do fp64 for that market, other formats for AI learning, targeting cryptocurrencies in ISA and data handling - how competitive is such a part likely to be on the discrete graphics market? Its like starting the tour de France on a monocycle, and then deciding to juggle four different balls as you race. Chances of competitiveness are slim. That's an exaggerated picture, of course, but trying to make a Jack of all trades chip when you are already at a disadvantage comes with costs in terms of engineering resources, time to market, die size and thus cost and power, harder work for the software side - the list goes on.
First off, being competitive doesn't
only mean having the best perf/area and/or perf/watt. Those two are great to have because you get to charge more (per-die-mm^2) than the competition for a product with similar performance, and make more profit out of it. But it's not the only thing that determines a product's competitiveness, there's also price and client focus, among other things.
As for the rest, I think AMD simply looked at it another way.
There isn't a single race. There are several races: gaming, CAD, compute, etc.
Instead of trying to desperately win first place in this single race called
gaming consumer (where they would hardly ever get a victory because of a substantial disadvantage in manufacturing process and R&D budget), they're making this jack-of-all-trades GPU that can run and
be competitive in almost all races.
So they're admittedly not winning the consumer gaming race, ok. But they're close second in many races and even winning a couple, like the Pro SSG in video editing. Remember the
client focus part? They had
RED's CEO going on stage saying "thank you for being the only ones to ever give a fuck", these were literally his words. And RED sells their own
video acceleration cards so Vega Pro SSG is likely to cut into the company's profits somehow -> this is how much they liked the attention (also, if professionals can work seamlessly with 8K, they'll be selling a lot more of their 8K cameras). Then there are the compute&AI cards that trade blows with GP100 (unless double precision is needed), CAD cards that trade blows with GP102, miners, etc.
As for Vega as a gaming card, I guess we're all a little
butthurt for us gamers not being AMD's primary focus (unlike every single card from AMD so far?). Take a few exceptions in this forum, we were all hoping for a card with Titan Xp performance while consuming 150W
(and while at it throw some way to hard-lock miners from using it).. and it's obviously not what we're getting.
But despite that, RX Vega
is competitive. Of course AMD is making less money on RX Vega than nvidia on GP104 at a similar MSRP, but they're definitely still making money on those cards and they're still competing. And lots of people will still prefer the RX Vega, whether because they purchased a Freesync monitor or they're willing to bet the card will largely surpass its current competition down the road, or something else.
In the end, given AMD's current inability to develop 5/6 different graphics chips at once every 18 months, releasing 1 or 2 chips per year that are either able to sell in several markets (Vega) or cater to a large segment of the gaming market (Polaris 10/11) is a much safer bet for profitable returns than putting all their eggs into the high-end gaming basket.
Even if that makes RX Vega look like it's racing the tour de France on a monocycle while juggling four different balls.
Does that mean we can see at some point a "refreshed" Vega with 14nm+ ? Is this even a thing at GlobalFoundries ?
Not according to
anandtech's latest article on GF's roadmap. Though it is indeed strange to see GF offering the exact same
flagship process for 2.5 years.
My guess is there's definitely some on-going process optimization happening between 2016 and 2018, and AMD probably knows better than anyone else the point at which they can call it "14nm+" and start making new chips that specifically take advantage of it.