While these specs aren’t officially released by AMD, a new set of rumors have popped up concerning AMD’s Radeon Fury series and they look quite accurate given recent benchmarks which confirm that the Fury X will indeed feature a full 4096 shaders/
AMD have already called the Fury the “world’s fastest GPU” and given the specs, it would certainly appear that this claim is fairly accurate. The GPU will of course feature High Bandwidth Memory (HBM) and will provide the GPU a rather tasty 512GB/s of memory bandwidth. This bandwidth should ensure that the GPU won’t run into bandwidth constraints and throughput will remain consistent.
There will be two versions of the GPU, both will feature the same amount of memory bandwidth – but the key differences lie in the number of shaders. The Fiji XT is the highest end part, and will clock in at an eye watering 4096 shaders, thanks to 64 GCN units. The GPU will also feature 128 ROPS and 256 Texture Mapping Units; it looks like AMD are taking their “built for 4k” motto rather seriously. The Fiji Pro is much the same, but features only 3584 Stream Processors, cutting the number of GCN Compute Units to only 56. You’ll still find 128 ROPS but TMU’s are cut to only 224. The XT also runs at a slightly higher clock speed, 1050Mhz compared to 1000.
Supposedly this means the top end Fury X will run at about 8.6 TFLOPS of performance, while the Pro will sit at around 7.2 – which is certainly no slouch, it’s still about 1.6 TFLOPS faster than the R9 290X. Of course, all of this also doesn’t take in to account of the various architecture improvements, as supposedly the Fiji XT architecture is an evolution of Tonga, meaning there will be improvements on tessellation and also better bandwidth efficiency thanks to color compression.
What does this mean for gamer’s? Well, at the very least the Fury X should (in theory) be about 50 percent faster than the R9 290x if we look at it from a pure TFLOP perspective, this doesn’t take into account bandwidth limited scenarios or various other architecture improvements.
Naturally, there will likely be some vendors who opt to run the Fury X faster, and speaking of which – there will be two iterations of the Fury X. One is a traditional air cooled GPU, featuring 3 fans, and the other is a closed loop water system featuring a radiator. As one can imagine, the later solution should mean the card is quieter, and (once again, in theory) mean you’ll be able to push for higher clock speeds.
|SPECS||Fury X (Water Cooled)||Fury X (Air Cooled)||Fury (Air Cooled)||R9 290X|
|GPU||Fiji XT||Fiji XT||Fiji Pro||Hawaii XT|
|GCN Compute Units||64||64||56||44|
|Render Output Units (ROPS)||128||128||128||64|
|Texture Mapping Units (TMU’s)||256||256||224||176|
|GPU Core Clock||At least 1050Mhz||1050Mhz||1000Mhz||1000Mhz|
|Memory||4GB HBM||4GB HBM||4GB HBM||4GB GDDR5|
|Effective Memory Speed||1Gbps||1Gbps||1Gbps||5Gbps|
|Cooling||Liquid, 120mm Radiator||Air, 3 Axial Fans||Air, 3 Axial Fans||Air, Single Blower Fan|
|Performance (SPFP)||≥ 8.6 TFLOPS||8.6 TFLOPS||7.2 TFLOPS||5.6 TFLOPS|