AMD Radeon HD 6900 Series: Right Performance, Right Features, Right Price

AMD Radeon HD 6900 series graphics cards
AMD is trying something new with its Radeon HD 6900 series of graphics cards. Instead of its usual tactic of addressing the higher end of the market with a graphics card containing two of the GPUs powering the Radeon HD 6870 (a chip code-named Barts), it is using a single, larger, more-powerful new GPU. Code-named Cayman, this new GPU doesn't merely take the architecture of the Radeon HD 6800 series and scale it up; rather, it makes significant architectural changes. AMD has redesigned the shader units for improved efficiency, greatly enhanced geometry performance, and employed a whole new power-management system.

The results are mixed. The Radeon HD 6900 series cards are certainly fast, though the 6970 doesn't always fare better than the Nvidia GeForce GTX 570 against which it is priced to compete. Anyone hoping that AMD would reclaim the absolute speed crown from Nvidia will be disappointed. Both the Radeon HD 6970 and the Radeon HD 6950 are quite long, too, and won't fit in smaller PCs (as is often the case with high-end enthusiast cards). Nevertheless, they offer good power efficiency for their class, as well as lots of other interesting features.

At almost 400 square millimeters, the GPU powering AMD's new graphics cards is the largest the company has produced in a long time, but it's still about 26 percent smaller than the GPU in Nvidia's latest cards. It features a lot of texturing power and high clock speeds, but fewer render back-ends than Nvidia's cards have. Note that the Radeon HD 6950 and 6970 feature 2GB of RAM, a step up from the 1GB we usually see on AMD cards; the extra RAM will help in intensive games running at high resolution.

Note, too, the discrepancy between the numbers of shader units for the AMD and Nvidia cards in the chart above; that discrepancy exists because the numbers are not directly comparable. Due to the different way the Nvidia and AMD chips are designed, a single shader unit from Nvidia is capable of doing more work than one in AMD's chip. An Nvidia shader unit is also larger, and therefore not as many of them are present in the GPU.

Next: A new architecture from AMD

A New Architecture From AMD

The architecture of the Cayman GPU is fairly new, with quite a few changes from that of the Radeon HD 5000 series and Radeon HD 6800 series. The chip features two "front-end" graphics engines, with two thread dispatch processors, two geometry tessellation units, and two vertex processing engines. This design should dramatically improve performance in geometry-intensive applications, particularly those that have a lot of tessellation. (Tessellation is a DirectX 11 feature that breaks up large polygons into lots of smaller polygons, to add detail or to smooth the edges of 3D models.) Here's a breakdown of some of the new features in the Cayman GPU found on the Radeon HD 6900 cards.

PowerTune technology: CPUs from Intel and AMD have a "turbo" feature that can crank up the clock speed a bit if the chip happens to be running beneath its rated thermal threshold. The new PowerTune feature of the Radeon HD 6900 cards sort of works like that, but in reverse: The core clock speed is set very high, and if the chip starts to run too hot, PowerTune lowers the clock speed in small steps until it's back at a safe level. Most graphics cards have to set the clock speed conservatively for the worst-case scenario of an application that strains the GPU and makes it run especially hot. You'll find a new slider in the Catalyst Control Center that lets you tweak the maximum thermal threshold at which PowerTune will kick in and slow down the card, from -20 percent to +20 percent. This will be an interesting function for overclocking enthusiasts to play with, but it has even bigger implications for the future, when this architecture eventually finds its way into AMD's "APUs" (combination CPU and GPU).

New antialiasing modes: In the Radeon HD 6800 series, AMD introduced a new antialiasing mode called Morphological AA. This post-processing effect antialiases everything on the screen, but can make the image appear slightly softer as a result. In the 6900 series, AMD has added Enhanced Quality AA. It's functionally similar to Nvidia's "CSAA" modes in that it adds a set of coverage samples to the usual number of color samples. This type of antialiasing can improve image quality with a pretty small impact on performance.

New VLIW4 shader architecture: The last several generations of AMD graphics chips used a VLIW5 architecture (that is, a five-way Very Large Instruction Word vector unit in each stream processing unit). One of the vector units in each set of five could handle special functions such as transcendental operations. Now, each vector unit is a four-way unit, with no special-function unit; if a transcendental operation is called, it instead occupies three of the four units. If all that sounds like a bunch of gobbledygook to you, you're not alone. The gist of it is that it should be easier for the GPU's scheduling hardware to keep all the math-processing units busy, providing for better performance in the same GPU area.

Enhanced geometry features: The Radeon HD 6900 series cards feature two graphics engines (the front-end processing of the graphics pipeline). That means two tessellation units, two geometry and vertex processing units, and two rasterizers. To make a long story short, the chip should be able to process a lot more geometry than previous AMD GPUs could. We still think that the high-end cards based on Nvidia's Fermi architecture, which also boast multiple geometry processing units, should offer somewhat higher overall geometry performance.

GPU computing enhancements: Along with reworking all those shader units from VLIW5 to VLIW4, AMD has made a number of other enhancements that should help with performance in applications that use the GPU for general computational tasks (such as physics, image post-processing, or video transcoding). The tweaks include improved flow control, dual direct memory access (DMA) units for faster memory reads and writes, and the coalescing of shader read operations. Most interesting is "asynchronous dispatch," or the ability of the GPU to work on multiple compute kernels simultaneously. In other words, you could ask the GPU to do some graphics work, some physics work, and some video transcoding, and the chip could split up those loads to its various parts to work on simultaneously, instead of switching back and forth among them. This function has limited utility today, but could be important in the future as the GPU becomes more of a shared system resource.

Beyond those changes, the new Radeon HD 6900 cards share the improvements that AMD made in the 6800 series: the UVD3 video engine, DisplayPort 1.2 and HDMI 1.4a support, per-display color correction, and HD3D (AMD's branding for 3D-display support).

Next: When the rubber hits the road

Performance: Synthetic Benchmarks

In our benchmarks, we compared AMD's Radeon HD 6970 and 6950--its two fastest single-GPU graphics cards--against Nvidia's two fastest single-GPU graphics cards, the GeForce GTX 570 and GeForce GTX 580. Note that the GTX 580 is a considerably more expensive card, at around $500. AMD says its competition for that product is the Radeon HD 5970, which is essentially two Radeon HD 5870 cards in CrossFire mode on a single, extralong card. The Radeon HD 6970 competes well against the GTX 570 (it should cost around $20 more than Nvidia's new card), and the 6950 model comes in $50 cheaper, at around $300.

We performed all our benchmarks on a system with an Intel Core i7 980X CPU, 6GB of RAM, and 64-bit Windows 7.

We started by taking a look at the Unigine Heaven benchmark, a synthetic test of a real DirectX 11 game engine, currently licensed by a number of smaller games. The test is rather strenuous and forward-looking, featuring high detail levels, dynamic lighting and shadows, and lots of tessellation. We run the test at the middle, "Normal" mode. In this tessellation- and geometry-heavy test, Nvidia used to dominate--but in this round, the Radeon HD 6970 outperformed the GTX 570. And at very high resolutions, the Radeon HD 6950 matched it.

FurMark is a synthetic OpenGL-based test that renders a torus covered in fur. It's fairly simple, but no test we know of stresses a GPU as thoroughly. It's a great way to see just how hot your graphics card will get and how much power it will use, but it isn't very useful as a real performance benchmark. On this test, Nvidia's power-draw safeguard kicked in and throttled down the GeForce GTX 580, seriously hindering its performance. The GTX 570--which draws less power and doesn't trigger the power safeguard--was actually able to outperform it. AMD's new cards ran this test very quickly, but it's difficult to tell just how much of a factor the PowerTune technology was. After we turned the maximum thermals up by 10 percent in the AMD control panel, we saw far higher performance in FurMark when antialiasing was disabled, but the numbers with antialiasing enabled didn't change at all.

New to our suite is 3DMark 11, a DirectX 11 test from FutureMark. It's a strenuous test that makes even the fastest graphics cards chug along slowly. Although it isn't indicative of current game performance, it may be a very forward-looking test that could clue us in on how these cards will perform in future games, relative to one another. In this test the Radeon HD 6970 was about as fast as the GeForce GTX 570 (slightly slower in the easier Performance mode, slightly speedier in the difficult Extreme mode). While the Radeon HD 6950 was about 12 percent slower than the 6970, it's still quite fast for a $300 graphics card.

Though it's getting a little long in the tooth, the DirectX 10 test 3DMark Vantage is still a common standard among synthetic graphics benchmarks. We present the 3DMark score with standard settings for the "High" and "Extreme" profiles. On this somewhat older test, AMD's new cards couldn't quite keep up with Nvidia's, even when we compared the similarly priced Radeon HD 6970 and GeForce GTX 570.

Next: Real game performance

Performance: Games

Synthetic tests can be useful in how they evaluate features that will be common in tomorrow's games, but performance in real games is far more important. We test with five current games that are all capable of pushing a modern graphics card to the limit.

Codemasters' Formula 1 racer F1 2010 is a more strenuous DirectX 11 driving game than Dirt 2, so we're moving forward with using it as a standard benchmark. We enable DirectX 11 and turn all the detail levels up to Ultra quality mode. AMD's cards looked great here, with the Radeon HD 6970 easily outpacing the GTX 570, and even matching the GTX 580 at extremely high resolutions.

Tom Clancy's H.A.W.X. is a graphically rich arcade flight game that utilizes DirectX 10.1 to enable features such as Screen Space Ambient Occlusion (SSAO), God Rays, and Soft Particles. Again, we turn all the detail levels up to the max here. Nvidia's cards easily took the prize in this game, though we should mention that all four of the cards managed over 60 frames per second even at extremely high resolution with antialiasing enabled.

Although World in Conflict is aging a bit, it's still a beautiful real-time strategy game with a DirectX 10-based graphics engine that can put a strain on all but the most powerful graphics cards when you maximize detail levels, as we do. The Radeon HD 6970 was roughly equal to the GeForce GTX 570 at the extreme resolution of 2560 by 1600, but Nvidia's new card was a bit faster at 1920 by 1200.

The S.T.A.L.K.E.R. series has always been on the bleeding edge of graphics technology. We use the demo benchmark for the Call of Pripyat sequel, with DirectX 11 lighting enabled and all detail settings maximized. The Radeon HD 6970 outpaced the GeForce GTX 570 by a small amount here--and at very high resolutions or with antialiasing enabled, the 6950 was just a step behind. Nothing could catch Nvidia's $500 GeForce GTX 580, but AMD's new cards came impressively close considering their lower price.

Last but not least, we have the excellent benchmark built into Just Cause 2. We maximize graphics settings and run the "Concrete Jungle" test, which is the most strenuous of benchmarks. We're beginning to sound like a broken record, but the pattern is becoming clear. In this game Nvidia's cards were definitely faster at 1920 by 1200, but AMD narrowed the gap considerably when we moved up to 2560 by 1600.

Next: Value and efficiency

Value and Efficiency

Looking over the performance of our test lineup, it's important to keep prices in mind. At a starting price of $300, the Radeon HD 6950 doesn't quite compete directly with the other cards in the group. The Radeon HD 6970 competes directly with the GeForce GTX 570; the two are separated in price by only $20. We included the GeForce GTX 580, priced at around $500, for reference--we thought it would be useful to compare AMD's fastest single GPU against Nvidia's fastest single GPU.

To find out which card offers the best value, we averaged the benchmark results for all our real-world game tests, and then divided by the price to arrive at a metric we call Dollars per Frames per Second. A lower number is better here; it means you have to spend less to get equivalent performance.

The GeForce GTX 570 looks to offer the best performance-to-dollars ratio at 1920 by 1200, but at the higher 2560 by 1600 resolution the Radeon HD 6950 provides value that is hard to beat. The Radeon HD 6970 doesn't offer quite the same ratio as the GTX 570, but it's still a far better deal than the GeForce GTX 580. Those $500 graphics cards are seldom good bargains; they're for price-be-damned enthusiasts who simply need the fastest thing out there.

Here's an impressive chart: total system power consumption. Although Nvidia greatly improved the power utilization of the Fermi architecture with the GTX 570 and 580, those cards still use a lot more power than the Radeon HD 6900 series cards do. In fact, the measured power came in well below the rated maximum power draw for both AMD cards; people who like to tinker with overclocking will probably find lots of headroom before things get too hot. Playing around with clock speeds and the new PowerTune slider in the control panel could yield impressive results, if these power readings are anything to go by.

By dividing the average frames per second of all our game tests by the power use under load from the previous chart, we arrive at a measure of Watts per Frames per Second. Instead of simply looking at how fast the cards are or how much power they use, this chart puts everything together to determine how power efficient they are. Here, again, lower numbers are better. At 1920 by 1200, AMD's new cards are very similar to Nivida's. Crank up the resolution to 2560 by 1600, and they hold a big advantage in performance per watt.

Next: Not a record breaker, but a good deal nonetheless

Not the Fastest Around, but a Great Total Package

The Radeon HD 6970 and 6950 are not record breakers. Fans of AMD's graphics cards may have been hoping that the Cayman architecture would dethrone Nvidia's GeForce GTX 580 for the single-GPU performance crown, but they'll be disappointed. It's not even close.

A great graphics card, however, is about more than just the highest frames per second. It's about delivering the right performance at the right price. Not everyone wants to pay $500 for a graphics card. For those willing to pay $350 or so, it's a tight race between the GeForce GTX 570 and the Radeon HD 6970. The GTX 570 probably offers slightly better performance in most games, unless you're using one of those big, 2560 by 1600, 30-inch monitors. On the other hand, the 6970 is more power efficient, and it features the ability to drive three monitors from a single card--a trick that Nvidia's cards can't pull off. It also offers a better array of display connectors, with two mini DisplayPort 1.2 plugs, two DVI connectors, and HDMI 1.4a.

So where does that leave the Radeon HD 6950? At $300, it may be our favorite card of the bunch. It's about 12 percent slower than the 6970 model, but 19 percent cheaper. It's about 15 percent slower than the GeForce GTX 570, as well as 15 percent cheaper, but it uses considerably less power and still has all those great display-connection features. Most of us are playing games on monitors with a resolution of 1680 by 1050, 1920 by 1080, or 1920 by 1200. At those resolutions, the Radeon HD 6950 offers enough performance to run today's most demanding games at maximum detail, usually with antialiasing enabled.

The best deals may be yet to come. Though the Radeon HD 6970 and 6950 both have 2GB of RAM today, AMD tells us that some of its partners will sell 1GB versions of the cards in the near future. The lower price tags on those cards could make them the best bargains around for people who don't have super-high-res monitors. If you have a monitor that's smaller than 24 inches, you might wish to wait for the 1GB versions before you upgrade. Until then, AMD has delivered a pair of very good graphics cards at compelling prices. They're not the fastest cards ever, but they're still quite speedy for the price, and they provide impressive power efficiency and display connectivity.

Subscribe to the Power Tips Newsletter