AMD Polaris Project F GPU Surfaces
On the linkedIn account from an ex-AMD employee there is a reference of a dGPU (Dedicated graphics card) tagged under codename F. The reference refers towards a GPU fabbed at GlobalFoundries and Samsung 14nm LPP. This could be the new mainstream AMD RAdeon GPU series.
The information also discloses that project F is a 232 mm² GPU (which is relatively small). Due to the die shrink on this new small node this could be the successor of the Pitcairn and Curacao which share roughly the same die-size hence my reference to mainstream.
However, when you apply some quick math, with the 14nm fab, this GPU could hold say roughly 4 to 5 billion transistors, which places it at a transistor level of the R9 380 en R9 380X in terms of perf (without taking into account the architecture changes).
According to AMD the Polaris architecture will be a mile-stone forward in performance per watt. Well, we can't wait ..
Junior Member
Posts: 15
Joined: 2015-12-22
Well, I would say something bit different:
AMD vs nV on 28nm. AMD always used bit higher density than nVidia. At cost of bit bigger leakage and lower clocks.
Fiji (Fury X) has 8.9B transistors in area of 596mm^2.
As can be seen , 14nm LPP takes is 3.55times smaller area than same block made on 28nm HPP.
But let's say AMD had it too dense last round around and this time they'll have lower density for improved power efficiency. And same amount of transistors will not fit 3.55 times smaller area, but will need more spare (2.5 times smaller only).
This "project F is a 232 mm^2" would have around "232 / 596 * 2.5 = 0.97" times as many transistors as Fiji. In other words, this is next Fiji!
I hope you are right man. Though every other tech news outlet I've read, ignores your math :/
Senior Member
Posts: 11808
Joined: 2012-07-20
Well, I would say something bit different:
AMD vs nV on 28nm. AMD always used bit higher density than nVidia. At cost of bit bigger leakage and lower clocks.
Fiji (Fury X) has 8.9B transistors in area of 596mm^2.
As can be seen here, 14nm LPP takes is 3.55times smaller area than same block made on 28nm HPP.
But let's say AMD had it too dense last round around and this time they'll have lower density for improved power efficiency. And same amount of transistors will not fit 3.55 times smaller area, but will need more spare (2.5 times smaller only).
This "project F is a 232 mm^2" would have around "232 / 596 * 2.5 = 0.97" times as many transistors as Fiji. In other words, this is next Fiji!