candle_86likely because its not competitive...
candle_86and AMD is trying to fix that
It's Hynix, the HBM 2.0 is causing the delay; maybe they used the time to work on drivers or keep raising the clock like they did with Ryzen, but the architecture is done, for better or for worse.
candle_86...otherwise will be fury X all over again, a decent card quickly forgotten as Nvidia steamrolls it with the next release
Think logically: Why bring something new ahead of competition? To maintain market share, otherwise they lose market share; nVidia isn't doing this for us, they do it for them.
It sucks that the consumer Vega is taking a while, but think about the fact that most people aren't loyal, they just buy whatever is available for a price they can bear. I'd wager they may not notice a "monopoly" unless either prices get higher or variety gets thin.
So far, nVidia has taken advantage of those that don't wait for better prices when competition finally comes around, it shows in their earnings reports. Those that waited for nVidia's preemptive strikes to consumer Vegas benefited by getting either a rebranded nVidia Titan-X 40% off, i.e. GTX1080 Ti, myself included
or a price drop in GTX1080 with faster memory-- both simply would not have existed if Vega wasn't near. Proof is CES 2017, nVidia didn't bring GTX1080 Ti then because Vega wasn't near enough for them to afford loosing premiums. It isn't greed, the point of running a business is making money.
candle_86...AMD has to stop being a product cycle behind NVidia.
Their parallel processing core architecture is some 60% behind nVidia, this is why they need more cores to line up with nVidia. For now, HBM X.0 gives them an edge where they can increase core count and up the clocks of their main GPU without crazy power draw that would result from GDDRX memory.
The Fiji die in Fury-X was 200W for itself at a base clock of 1GHz, add in HBM and other circuitry and the total was 275W. Shrinking the Fiji down to 14nm from 28nm would have meant cutting the power in half, and keeping everything else means 175W if at 1GHz. Raising the shrunk FIji to 1.6GHz would make the total 235W or so. Against a 220W GTX1080 Ti, it is within striking distance. But if AMD used GDDR5X on Vega-10 just to get the card out sooner, it would be over 300W, it would be a joke.
The oddity to me is the FP16 performance of Volta, it isn't that much faster than Pascal, 20.6 TFLOPS versus 30 TFLOPS. Granted, it scales almost equally with extra CUDA cores and the clocks, but Vega at the same clock is already 25 TFLOPS-- this is why Volta came at all, to fight the professional Vega (card in the article) which is actually somehow faster than consumer Vega.