Wednesday, July 22, 2015

AMD Radeon R9 Fury X

The R9 Fury X is AMD's new flagship graphics card. It features a brand new GPU codenamed Fiji and it's also the first card to sport high-bandwidth memory (HBM). Liquid-cooled as standard, it's set to retail for just over $800, putting in direct competition with Nvidia GTX 980 Ti.

Fiji is essentially an extended version of the Hawaii GPU used in AMD's R9 290 series. It uses the same Graphics Core Next architecture, but each of the four shader engines that make up the bulk of the pipeline now feature 16 Compute Units rather than 11, for a total of 4,096 stream processors and 256 texture units - an increase of 45 per cent over its immediate predecessors.

As such, the transistor count has increased to a massive 8.9 billion, with a die size of 596mm²; after all, its still made on the same TSMC 28nm process as Hawaii. The Fiji GPU is also clocked at up to 1,050MHz; AMD's quoted clock speeds are the maximum at which a GPU will operate within its thermal limits, but thanks to the liquid cooler, throttling won't be an issue with the Fury X.

Meanwhile, the GPU's front and back end are unchanged. At the front, the graphics command processors is paired with eight asynchronous compute engines, so graphics and compute tasks can be handled simultaneously. At the back, AMD has stuck to 16 render back-end units (four per shader engine), thus keeping the ROP count at 64.

Other features of this latest GCN architecture include TrueAudio, AMD's dedicated sound processor for in-game positional audio. There's also CrossFire XDMA, which means multi-GPU configurations can communicate directly over the PCI-E3 bus, with no external bridge connector. As you would expect, the R9 Fury X supports DirectX 12 too, although not the full 12_1 feature level supported by the GTX 980 Ti. Still, it supports the major enhancements, including async shaders as well as the optimizations made for multi-GPU setups and multi-core CPU's

Sponsored Products Related To This Article

The real newsworthy change, though, is the move from GDDR5 to HBM, as GDDR5 is too power-hungry at high clock speeds, limiting the power budget available to keep GPU performance high. GDDR5 memory also requires a lot of PCB space when you need multiple GDDR5 dies surrounding the GPU.

HBM directly address these issues. It uses smaller dies that, thank to their extreme thinness, can be stacked on top of one another. They can also be placed much close to the GPU using a layer of silicon called an interposer, through which the GPU and DRAM can be directly connected. This setup results in significantly lower power consumption, and the ability to use much wider interface for greater bandwidth. HBM will also be adopted by Nvidia when its Pascal architecture launches.

The stacked dies are connected using a technology called through-silicon vias, or TSVs, which are tiny holes drilled through the silicon, filled with copper and joined with solder microbumps. This first generation of HBM is limited to 256MB dies and four dies per stack. The fiji GPU can communicate with four stacks, so the R9 Fury X is limited to 4GB of VRAM, but AMD says that the increased bandwidth means the card still has the memory performance needed for high-detail 4K gaming.

Each stack has 1,024-bits of bandwidth via eight 128-bit channels (two per die) and, as Fiji has eight 512-bit memory controllers (two for each stack), the end result is a massive 4,096-bit wide interface. With such a wide interface, clock speeds can be drastically lowered, saving more power. Fiji's HBM is clocked at 500MHz, and thank to its double data rate, its effective clock speed is 1GHz, giving it a total available bandwidth of 512GB/sec - far more than any other card around.

The use of HBM also means the total space occupied by the GPU and 4GB of VRAM is roughly three times smaller than on the R9 290X. Consequently, AMD has managed to cram all the parts onto a much smaller PCB - it measures less than 200mm long, although you have to contend with the 120mm radiator too.

The Cooler Master cooler makes contact with the GPU and HBM stacks via a copper baseplate. The tubing from the pump unit then passes through a copper heatpipe, which is soldered to a metal contact plate that draws heat away from the VRMs in the 6+2 phase power design, negating the need for a fan on the card itself. Meanwhile, the exposed tubing is 400mm long with high-quality sleeving, and the 120mm radiator ensures the best possible case compatibility.

AMD has clearly learned some lessons since the R9 290X too, as the card's build quality is great. The all-metal card merges black nickel-plated sections with soft-touch aluminum part. The front faceplate can also easily be removed and replaced if users (or third parties) want to 3D print or make their own designs. There's a red LED behind the top Radeon logo as well, and just above the two 8-pin PCI-E connections is a set of eight LEDs that light up to indicate the current GPU load. These LEDs can also be set to blue, and there's a separate green LED to tell you when the GPU has entered its ZeroCore power state. Lastly, the Fury X has a dual BIOS switch, so you can have a stock speed and overclocked profile if you wish.

In term of outputs, DVI support has been dropped - instead, the Fury X has a trio of DisplayPort 1.2 connectors with FreeSync support and a single HDMI 1.4a socket. The latter is an odd choice, though, as it means the card can't be used with the vast majority of 4K 60Hz TVs until DP1.2 to HDMI 2.0 adapters become available later this year.


At 2,560x1,440, the GTX 980 Ti is better in every test, with its biggest leads seen in Battlefield 4 and The Watcher 3. In other games, though, it leads by only a few frames per second. In many cases, you're unlikely to notice the difference, but many people will have been hoping for a better result from AMD's new best card. Only in two games at this resolution does the Fury X stay above 60fps, whereas the GTX 980 Ti manages this achievement in four of them. Overall, Nvidia's card is faster by 8 per cent on minimum frame rates, while the Fury X improves over the 290X by 35 per cent.

AMD Radeon R9 FuryX benchmark test1

AMD Radeon R9 FuryX benchmark test2

AMD Radeon R9 FuryX benchmark test3

At 4K, though, that memory bandwidth comes into play and the Fury X is much more competitive; the 4GB of VRAM really doesn't appear to be a limit. In all our game tests performance is practically identical.

Like the GTX 980 Ti, the Fury X offers playable frame rates in all games, although its Crysis 3 frame rate is right on the borderline of playability. At this resolution, it's faster than R9 290X by 38 per cent.

AMD Radeon R9 FuryX benchmark test4

AMD has also reined in its power consumption, with our system only consuming 34W more power with the R9 Fury X installed, compared with the GTX 980 Ti, despite neck and neck performance at 4K. Temperature and noise are two more strong points. There's some slight pump noise and coil whine that's audible, but the fan is virtually inaudible even after prolonged load. The GPU temperature also peaked at just under 60°C while the fan stayed at its minimum speed of 15 per cent.

AMD Radeon R9 FuryX benchmark test5

The Fury X isn't a great overclocker though. AMD says it's removed thermal and electrical limits by using a powerful liquid cooler and beefing up the power delivery, but the GPU core frequency still topped out at 1,130MHz, a boost of less than 8 per cent. The HBM can't be overclocked either, nor could we increase the voltage. The end result is only very small increases in performance - far less than what you can achieve with a GTX 980Ti. On the plus side, power consumption only increases slightly and there's no noticeable change to heat or noise output.


The AMD Radeon R9 Fury X is a solid, well-rounded card in many respects, but it isn't without flaws, and it's not the Nvidia-crushing beast for which many had hoped either. The missing HDMI 2.0 support is an oversight, and its DirectX 12 support isn't fully featured either. Lastly, the Fury X is a pretty disappointing overclocker, operating at pretty much maximum speed out of the box.

AMD Radeon R9 FuryX benchmark test6

On the other hand, it's really well made, exceptionally cool and quiet, and has lower power consumption than we expected - you can easily run it in a high-end overclocked system that draws less than 500W from the main. It's also pleasingly small - of course, much of the size is transferred to the radiator, but plenty of mini-ITX cases can still handle a 120mm radiator. The middling 2,560x1,440 performance is a shame, but at 4K, it's competitive enough to be award-worthy. That said, it faces a close battle with Nvidia's GeForce GTX 980 Ti, and it you can get a good deal o a custom-cooled, overclocked GTX 980 Ti, you'll end up with a better card.


Graphics Processor: AMD Radeon R( Fury X, up to 1,050MHz
Pipeline: 4,096 stream processors, 64ROPs
Memory: 4GB 1GHz (effective) HBM, 4,096-bit wide interface
Bandwidth: 512GB/sec
Compatibility: DirectX 12, Mantle, Vulcan
Inputs/Outputs: 3x DisplayPort, HDMI
Power connections: 2x 8-pin, top-mounted
Size: 195mm long, dual slot
Radiator: 120mm (400mm tubing)

By Matthew Lambert

Sponsored Products Related To This Article
PC matter, learn about your pc needs, software and peripheral devices
Share This
Subscribe Here


Post a Comment