I believe this was written by
@btarunr .
Man I don't fully understand AMD, their GCN architecture is finally showing it's might with the introduction of DX12 and Vulcan and then AMD goes and makes an entirely new architecture. Will this new NCU engine require 4 years to reach it's potential? Does the consoles have this architecture? Maybe Scorpio? I hope so for AMD's sake.
From page 1, first paragraph:
The GPUs will still employ GCN compute units
the big news here is the way the units are fed, AMD has added more functions and hardware to ensure that all more of the units are ready to work on more data.
This is logical, seeing as Fury has an high theoretical performance from the GCN compute units, but not always that much more performance.
They basically trying to make a GPU prefetcher.
My concern is all the reference to the drivers to help here, meaning that there will be room for performance improvements when the card finally is released.
It seems AMD is getting ready for a big push into the AI market with Zen and Vega. Perhaps since the money is there and not in super low margin OEM APU sales they will come out of their decade long slump.
About half the Vega features are AI and compute technologies being sold to us gamers as vaporware, don't get me wrong if it works for the price great, but the features they are touting are designed and made more for compute than graphics.
Yeah, this is a one size fits all chip, i do not expect it to be as efficient as GP104 (unless AMD has done some magic here) since that chip is a gaming chip through and through, but i might be wrong.
