AMD Radeon R9 Fury X tested: not quite a 980 Ti killer

h1

The AMD R9 Fury X is not the 4K champion, 980 Ti killer AMD would like it to be. Based on my time testing this new graphics card with high bandwidth memory, it’s fast— sometimes neck-and-neck with Nvidia’s equally priced 980 Ti—but not quite fast enough, often falling about 10 percent short of Nvidia’s card in average framerates, across a range of games.

This isn’t a full, comprehensive review of the Fury X, because I haven’t had enough time with the card to provide a complete analysis (thanks, E3!). I’ve had a full day’s testing to put the card through its paces at 1080p, 1440p and 2160p in a number of game benchmarks. It’s enough testing to have a confident overview of the Fury X’s performance and how it stacks up against the competition, minus one important element: overclocking.

For starters, here’s the usual spec breakdown of the R9 Fury X and its major competitors.

h2

Now, then. Ready for some charts? I know you’re ready for some charts. The R9 Fury X benchmarks below were performed on the following Windows 8.1 system, using AMD’s 15.15 drivers:

 Intel Core i7-5960X, clocked at 4.2GHz with 2 cores disabled
 16GB DDR4-2666 RAM
 ASRock X99 OC Formula motherboard
 1TB Samsung 850 Pro SSD

These specs were meant to mirror, as closely as possible, the test bench of our colleagues over at Maximum PC. Their rig uses an i7-5930k processor at 4.2GHz (hexacore compared to the 5960X’s octocore, hence our two disabled cores) and a Gigabyte Gigabyte GA-X99-UD4. Due to some slight differences between the motherboards, we did notice a small performance gap between the two rigs when comparing Titan X framerates at 1080p, when games are more likely to be CPU-bound than GPU-bound. However, at 1440p and 4K, those differences disappeared, delivering identical average framerates across several games in our benchmark suite. After a full day of scrutinizing the performance, running and re-running benchmarks, we’re confident that the R9 Fury X numbers here can be accurately compared against our previous testing.

R9 Fury X benchmarks: average framerates

If you factor in a small margin for error (and the variance that comes from one benchmark run to another), the R9 Fury X performed very closely to Nvidia’s 980 Ti, its closest competitor. However, in our benchmarks, which always used Ultra settings and enabled 4xMSAA when possible, the Fury X regularly came up short of the 980 Ti and Titan X, both of which offer more VRAM.

This contrasts with the numbers AMD has provided for the card, which show the Fury X just barely outperforming the 980 Ti in a dozen benchmarks. Those benchmarks, however, were run with varying settings on a game-by-game basis, like medium, FXAA in Assassin’s Creed Unity but Ultra, 2xSSAA in Sniper Elite 3. Without fine-tuning settings for each game, we didn’t see the same advantage for the R9 Fury X.

Update: I forgot to point out that our comparisons to the 980 Ti below are using the factory over clocked EVGA 980 Ti, which had a base clock of 1102 Mhz and boost clock of 1190 MHz. That accounts for part of the discrepancy between the cards, as well as recent driver updates on Nvidia's part that improved performance. The Titan X is running at stock factory speeds, comparable to the 980 Ti. If you take away the advantage of the overclock, the cards do turn in very similar performance, as you can see in some cases comparing the Titan X to the R9 Fury X, but it's rare for the Fury X to pull ahead. I apologize for not being clearer, and will make sure to update with overclock numbers for the Fury X when I have them. 

h3

In Arkham Origins, an Nvidia-optimized game, you can see that the Fury X just barely underperforms the Titan X and 980 Ti Superclocked, which turn in equal scores.

h4

h5Hitman and Tomb Raider are both Square Enix games, and typically perform extremely well on AMD cards. Hitman is the only game in our testing that saw the Fury X outperform the Titan X and the 980 Ti SC, but it didn’t fare quite so well in Tomb Raider. Still, it delivered an impressive 43.8 fps at maxed-out Ultra settings at 4K resolution.

h6Here it’s worth pointing out that the Fury X just barely outperformed the GTX 980. We’ve seen recent driver updates on Nvidia’s side give a nice little performance bump to the 980 Ti SC and Titan X; those framerates were on the 353.30 driver, giving both cards a significant fps lead over the Fury X.

h7

The Fury X fared well in The Witcher 3 with Hairworks disabled, though again it lagged a bit behind the other two cards. However, as The Witcher 3 doesn’t contain a benchmarking tool, we’d expect a bit wider margin of error in these scores, putting the three cards in a fairly even heat.

R9 Fury X benchmarks: minimum framerates

While average framerates give you a good overview of general performance, minimum framerates alert you to potential game-ruining performance pitfalls. Nasty framerate drops can ruin your gaming experience, even if they don’t happen too often. While running in-game benchmarks, we didn’t notice any serious issues with smoothness or dropped frames with the Fury X. But much like average framerates, we saw the Titan X and 980 Ti SC regularly outperform the R9 Fury X. Here, usually by a slightly bigger margin.

Look at the smaller bars in the charts above for that comparison.

h8

4GB of High bandwidth memory and AMD’s drivers

The R9 Fury and the watercooled Fury X are groundbreaking releases for AMD, using the new high-bandwidth memory in place of GDDR5. It’s a promising technology that’s only going to get more and more impressive and important with future iterations, but HBM 1.0 in use in these cards brings with it a problematic limitation: 4GB of memory. Where Nvidia’s 980 Ti and Titan X offer 6GB and 12GB of VRAM, the Fury X is stuck at 4GB. While this isn’t a problem for gaming at 1080p, it may soon be a problem at 1440p, and it’s definitely already a limitation at 4K.

Grand Theft Auto 5, for example, is able to use up the 980 Ti’s entire 6GB of VRAM with MSAA and its advanced settings enabled. When we ran GTA5 totally maxed out on the Fury X, 980 Ti and Titan X, the smaller memory pool ended up giving the Fury X an unplayable minimum framerate of 6 fps compared to 17 fps and 22 fps for the Nvidia cards.

Granted, those are extremely demanding settings, and with some proper tweaking, you can certainly get playable framerates out of the Fury X. At 4K, with no advanced settings and no AA, the card turned in an average 46 fps, virtually matching the Titan X.

But we’re worried about the Fury X being future-proof—the card is a significant investment at $650, and there’s a real possibility that games in 2016 and 2017 will routinely guzzle six gigabyte frame buffers. For 1080p gaming, the Fury X is a very strong card, but it’s not notably stronger than the 980 Ti, which nets you two more gigabytes of memory for the same price. That’s a hard sell.

h9

Overclocking, liquid cooling and conclusion

The Fury X’s framerates are tantalizingly close to the 980 Ti SC's, and it’s hard to know exactly what’s holding it back: the hardware of the card itself, or AMD’s drivers. Given the 980 Ti SC's factory overclock, that puts the stock 980 Ti at about equal performance, but with its RAM advantage. But Nvidia's recent drivers have given the 980 Ti some nice performance bumps; AMD’s been slow to release optimized drivers for years, and it’s quite possible that the Fury X is launching with poorly optimized drivers. An update could, potentially, pull the Fury X dead even with its competition, and nudge games like Tomb Raider and Hitman Absolution into the win column for AMD.

But that’s speculation. For now, Nvidia’s more frequent driver updates and better software experience make their cards more compelling.

The Fury X’s one other hope is its liquid cooling potential. The card’s amazing 7.5-inch length is possible thanks to the new HBM memory a beefy external radiator, which may make it difficult to install in some cases—that’s the drawback of its fantastic cooling potential, which idles at about 28C. While the Fury X draws less than 300W at a standard load, AMD claims it has headroom for up to 375W. If you're building a rig that won't have space for the liquid cooled radiator, the air-cooled Fury will be the same size, minus the radiator, and there's an even smaller (lower-TDP) Nano card coming later this year.

We haven’t had the opportunity to explore overclocking performance yet, but even there the air-cooled 980 Ti will put up some strong competition. The 980 Ti can increase performance by a good 20 percent overclocked, so the Fury X’s overclocking will have to be truly extraordinary to surpass it.

Until we’ve explored the overclocking potential of the Fury X, our recommendation is to wait on a purchase. Drivers could improve performance, and the air-cooled R9 Fury will also be available at a lower price point. That may make its 4GB of memory more palatable. But from what we’ve seen, the Fury X doesn’t significantly outperform the 980 Ti, and Nvidia’s card offers 2GB more VRAM and better drivers for the same price.