There have been plenty of talk regarding AMD's newest flagship Radeon Fury lineup and it was about time we get more details regarding the actual GPU behind it, which is the largest GPU ever made by AMD.

Although we have already heard most of the specifications even before it was launched, including the fact that it packs 4096 Stream Processors, 256 TMUs and 64 ROPs, as well as feature 4GB of High Bandwidth Memory with 4096-bit memory interface, stacked on the same package via interposer, we did not have a chance to hear precise details regarding the GPU itself.

In case you have been wondering, AMD's new Fiji GPU packs 8.9 billion transistors on a die are of 596 mm2, which makes it the largest GPU AMD ever made, and this number does not include HBM and the interposer area. If you take those numbers into account, the die area, or to be precise, interposer area, is 1011 mm2 with over 10 billion transistors.




These numbers make the Fiji GPU almost as big as the Nvidia's 28nm GM200 GPU, which has a die area of 601 mm2 and packs 8.0 billion transistors.

When it comes to block diagram, Fiji GPU has 4 Shader Engines, each with 16 Compute Units, with each Compute Unit packing 64 Stream Processors, for a total of 4096 Stream Processors. It has eight memory controllers, which are connected to four HBM memory stacks in pair.




Compared to the Radeon R9 290X Hawaii XT GPU based graphics card, the PCB area occupied by ASIC and memory is 3x smaller, thus allowing AMD to make the Fury lineup shorter than what you would expect from a high-end graphics card.




AMD's Fury lineup and the Fiji GPU look quite impressive and all we need to see now are some game benchmarks.



Source: Videocardz.com.