| thismeintiel said: Also, Navi has been put on hold til 2019. |
That is only a Rumor. Untill such a time AMD has confirmed it, we should trust AMD's roadmap to some degree.
| thismeintiel said: Is Zen and Vega better than what AMD has, now? Yes. |
Well. That couldn't be anymore obvious.
But just because it's better than what AMD has now, doesn't mean the goalpost won't be shifting over the next several years.
| thismeintiel said: And I'm sure they will be using some improvements that AMD develops along the way, just as the PS4 Pro GPU does. Using somewhat older tech is just something console gamers are used to, considering these things are launching at just $399. |
The Playstation 4 is an old GPU design. It was old when it released. Having a few "customizations" didn't dramatically change it's capabilities... And the games speak for themselves.
In 2019/2020, mainstream hardware will not only be more efficient than AMD's first Generation NCU architecture, but also cheaper.
| thismeintiel said: Edit: I also know that AMD/GF is switching to 7nm. I don't see how that refutes a Vega GPU, considering they are supposed to release a Vega made using 7nm in early 2018. |
7nm is not going to arrive in early 2018 for large monolithic chips at Global Foundries.
It will arrive in late 2018, early 2019. Which may push next-gen consoles releasing to 2020.
But don't take my word for it: globalfoundries.com/newsroom/press-releases/2016/09/14/globalfoundries-extends-roadmap-to-deliver-industry-s-leading-performance-offering-of-7nm-finfet-technology
2018 RISK production starts in early 2018. Commercial chips come much much much much later.
So no. You will not get Vega in early 2018 built at 7nm.
Heck, Global Foundries has more of a history of cancelling/delaying it's new nodes than actually being on time anyway. So don't be surprised if it blows out to late 2019.
New process improvements are also slowing down and per-transister costs are increasing due to heavier reliance on things like multi-patterning, EUV and other technologies.
Samsung will release it's 4th Generation 14nm process soon (If it hasn't already) which should be fairly competitive for awhile to come.
If 7nm flops at Global Foundries, then they will likely just license Samsungs technology again, Samsungs 10nm process will be a Hybrid process where the BEOL or FEOL is 14nm (Some rumours are saying 20nm).
TSMC's 10nm process should have a density advantage over Samsung with some luck. (Some estimates place it at near Global Foundries 7nm in terms of density.)
And then you come to the conclusion that the fabs naming schemes for process shrinks is bullshit and not comparable.
www.youtube.com/@Pemalite
