The AMD Fiji Pro GPU

AMD Fiji

So, AMD likes to stick to island series GPUs for their product development codenames, for the Fury series the Fiji GPU gets its denomination from there. AMD Fiji is the new GPU used, we are going to talk about the GPU now and through part of the next page we'll talk a little more in-depth about the HBM that sits on that GPU. The Fiji series of GPUs is the next iteration of the GCN (Graphics Core next) architecture. Fiji is based on revision 1.2 GCN, just like Tonga (285 & 380 series). Basically, if we took Tonga and scale that upwards that would be Fiji (but that is an overly generalized explanation).

	Radeon R9 Fury X	Radeon R9 Fury	Radeon R9 390X
Fabrication Process	28nm	28nm	28nm
GPU	Fiji	Fiji	Hawaii / Grenada
Streaming Processors	4096	3584	2816
Graphics memory	4 GB HBM	4 GB HBM	8 GB GDDR5
Memory Clock	up-to 500 MHz / 1.0 Gbps	up-to 500 MHz / 1.0 Gbps	6.0 Gbps
Core Clock	1050 MHz	1000 MHz	1050 MHz
Memory Bandwidth	up-to 512 GB/s	up-to 512 GB/s	384 GB/s
Power Connectors	2 x 8-pin	2 x 8-pin	1 x 6-pin - 1 x 8-pin
Form Factor	Full Height, Dual slot	Full Height, Dual slot	Full Height, Dual slot
Freesync	Yes	Yes	Yes
DirectX 12 Support	Yes	Yes	Yes

Fiji is still based on a 28nm fabrication node, 20nm failed and until we see 16 and 14nm viable for big GPUs, the 28nm node really is the only alternative. That was the same dilemma for Nvidia with the GTX 980 Ti of course. Both GPUs as a result are big. For Fiji that Mac Daddy is 5x5cm in size. Bigger means higher voltage usage, and thus more heat to deal with. You can only imagine what a chip like this would be like on 14nm, at half the size. AMD did huge things with Fiji though, they tucked away 8.9 Billion transistors on the GPU and that's the number without the memory. It gives them a staggering 8.6 TFLOPS of performance to work with. That, combined with HBM memory, is to be considered serious stuff. The GPU has a whopping amount of transistors. At 28nm AMD applied the GCN 1.2 architecture, albeit you will learn that there have been a few tweaks at cache levels as the L2 cache is now 2MB for example.
 

Radeon	R9 Fury X	R9 Fury	R9 390X
Process	28nm	28nm	28nm
Transistors	8.9 Billion	8.9 Billion	6.2 Billion
Engine Clock	1050 MHz	1050 MHz	1050 MHz
Stream Processors	4096	3584	2816
Compute Performance	8.6 TFLOPS	7.2 TFLOPS	5.9 TFLOPS
Texture Units	256	224	176
Texture Fillrate	268.8 GT/s	224.0 GT/s	176 GT/s
ROPs	64	64	64
Pixel Fillrate	up-to 68.2 Gp/s	up-to 64 Gp/s	up-to 64 Gp/s
Z/Stencil	256	256	256
Memory Bit-Interface	4096-bit	4096-bit	512-bit
Memory Type	4GB HBM	4GB HBM	8GB GDDR5
Data Rate	500 MHz / 1.0 Gbps	500 MHz / 1.0 Gbps	1500 MHz / 6.0 Gbps
Memory Bandwidth	512.0 GB/s	512.0 GB/s	320.0 GB/s
Typical Board Power	275W	275W	275W
PCI-Express Gen	3.0	3.0	3.0
API support	DirectX 12 / Vulkan / Mantle	DirectX 12 / Vulkan / Mantle	DirectX 12 / Vulkan / Mantle

An interesting step from AMD was  the move towards a 4096-bit memory bus thanks to HBM. Nobody really expected it as in the past AMD merely placed GDDR on their products. You will notice very similar structures compared to Tonga (Radeon R9 265 and 380). This is GCN 1.2, yet multitudes of it. The Fiji XT GPU as such has 4096 shader processors / and Fiji Pro 3584 shader proceccors, these are clustered up in groups of 64 / 56  shader processors per CU (compute unit).