AMD Radeon R9 Fury - Review

Last week at E3, AMD pulled the wraps off its new flagship GPU: the liquid-cooled Radeon R9 Fury X. It’s the most powerful GPU AMD has ever created, as well as the company’s first video card to feature new architecture and a new type of memory. Even before its announcement, speculation ran wild about how it’d compare to Nvidia’s GTX 980 Ti, which is equally priced at $650.

When specs and AMD-provided performance numbers came out late last week, it appeared that the two cards were pretty closely matched—and after some extensive testing, we can confirm that both cards are indeed pretty darn close in performance.

While the Fury X performs about the same as the GTX 980 Ti, the approach that AMD took with it is quite different in some ways. Let's dig in to its features.

High Bandwidth Memory

The biggest new feature in the Fury X is one that's never appeared on a consumer video card previously: its use of high-bandwidth memory, or HBM for short. This technology eschews individual RAM modules distributed around the GPU that connect via wire traces; instead, it uses vertical stacks of memory placed right next to the GPU die. This saves a lot of space on the GPU, since the memory sits on top of the GPU package. Also, the interposer that connects the two allows for a much wider bus, which lets AMD run the memory at lower clock speeds.

The result is a massive improvement in memory bandwidth that takes up much less space and consumes far less power than earlier types of memory. To put it into numbers, the Fury X’s memory is clocked at just 500MHz and delivers 512GB/s of bandwidth over a bus that is 4,096-bit wide. The GTX 980 Ti’s memory is clocked at 1,753MHz (quad-pumped to 7GHz), and yet it only delivers 336.5GB/s of bandwidth over a 384-bit bus. Pretty big difference, no? This technology is so badass Nvidia has already said that HBM will appear in upcoming GPUs, but you won’t see it until the company’s Pascal architecture arrives sometime in 2016.

Fiji

The Fury X uses an all-new chunk of silicon dubbed Fiji, as opposed to the die in the Radeon R9 290X, which was called Hawaii. Fiji is the beefiest GPU AMD has ever crafted, and it sports a staggering 8.9 billion transistors. To put that number in perspective, the GM200 at the heart of the GTX 980 Ti has 8 billion transistors, while the R9 290X has 6.3 billion. This particular video card’s also packing a class-leading 4,096 stream processors, and 4GB of the aforementioned HBM.

Liquid-Cooling

Unlike previous liquid-cooled GPUs that relied on two fans—one on the card to cool the components and another to cool the radiator—the Fury X’s system cools everything with just one 120mm fan. The cooler utilizes copper pipes that make contact with a heat spreader that sits atop the board’s components. The pipes absorb the heat and then pass that along to the water as it moves through. There’s also a small reservoir located beneath the radiator, allowing more coolant to be used in an effort to improve performance. AMD even braided the tubing, which is a nice touch. AMD claims this radiator keeps the GPU around 50°C under load, which bore out in our tests; it’s also incredibly quiet. Its Nidec-produced Gentle Typhoon fan only generates 32db of noise. (That’s quieter than being inside a library.)

GPU Tach

AMD has placed 9 LEDs on the back of the GPU that function as a “tachometer” indicating the current level of GPU utilization. Eight of them fully light up when gaming, and it does look neat. By default they’re red, but you can change the color to blue (that seems like blasphemy to us, though). There’s also one green LED that glows when the GPU is in ZeroCore power-saving mode, but we never saw it come on during testing. You can also turn the lights off completely.

Other Features

In addition to everything above, this flagship GPU supports AMD’s FreeSync technology, which synchronizes the refresh rate of the display to the frame rate of the GPU to eliminate tearing; DX12 and Mantle; Virtual Super Resolution, which is AMD’s version of downsampling; and Frame Rate Targeting, which lets you set a frame limit in order to conserve energy. AMD also says the Fury X will be awesome when used with unreleased VR equipment, but we’ll have to wait and see about that.

Performance

All tests were run on our test bench running an Asus Z97-A motherboard, Intel Core i7-4790k motherboard, 16GB of DDR3/1600 memory, 240GB Crucial M500 SSD, 1000W Corsair PSU, and Windows 8.1. All tests at 1920x1080 and 2560x1440 were run at maximum graphical settings with 4XAA enabled, with the exception of Batman: Arkham Origins, which had PhysX turned down to Normal to allow for comparable tests between the AMD and Nvidia cards. Tests at 3840x2160 did not have AA enabled.

Like the rest of humanity, we were very eager to fire up the Fury X to see if it was faster than a GTX 980 Ti, as well as how well its liquid-cooling mechanism functioned. To answer the first question, yes, it is as fast as a GTX 980 Ti, but not really faster. At 4K resolution, we saw a draw in three out of the six games we tested. In the remaining three games Nvidia was faster in two and AMD faster in one, making it a nail-biter. We saw the same scenario at 2560x1440 too, though the 980 Ti held a decisive lead in five out of six games at 1080p. That last point shouldn't matter to buyers of this card in practice, though—you won't be getting it for 1080p gaming.

Overall, no card dominated these proceedings, but instead the two matched each other quite well throughout testing. You could say Nvidia holds the advantage simply because it did all this with air-cooling, while AMD with its liquid setup and HBM only just matched it. However, to AMD’s credit, the Fury X is totally silent and runs a lot cooler than the 980 Ti (which hits 83°C), so again, it’s a back-and-forth thing with no clear winner.

Overclocking

AMD has marketed the Fury X as being designed for overclocking—and since it’s water-cooled and runs at just 50°C, you would think there’s a lot of headroom for overclocking…but we had no luck with it. In the documentation that AMD shared, it says that a 100Mhz overclock netted about 5% gains in a few games such as Far Cry 4. That’s nothing to write home about, especially considering Nvidia’s Maxwell GPUs clock up to about 1.4GHz with relative ease. We tried using AMD’s PowerTune and got a 6% overclock out of it, but that’s not impressive; it only yielded about one or two fps more in games. When we asked AMD about it, we were told that the amount of overclocking possible will vary from card to card, and so it could not offer a general range one could expect—which is disappointing on such a well-cooled GPU.

We also discovered AMD has removed the ability to tweak the memory clocks, but it’s probably not necessary since the memory likely isn’t a bottleneck.