AMD introduced two new graphics cards for the enthusiast segment on December 15, 2010. Known as Radeon HD 6970 and Radeon HD 6950, this TekSpek examines the new architecture, dubbed Cayman, and then compares it with NVIDIA's Fermi core, which powers the recently released GeForce GTX 580 and GeForce GTX 570 cards.
Cayman architecture
The guts of the two new Radeon cards are based on what AMD calls the Cayman core. This is an improvement over the Cypress core that powers the year-old Radeon HD 5870 and the two-month-old Barts core found in Radeon HD 6870 GPU.
The Cayman core represents the most complex graphics processing unit (GPU) AMD has ever designed for the gamer. Packed with 2.64bn transistors and taking up 389mm² of silicon, the new GPU provides double the geometry-processing power of the last-generation Radeon HD 5870. AMD has decided to focus on boosting geometry because it believes, like NVIDIA, that it's the cornerstone of future games development.
Geometry is increased by doubling what are termed the graphics engines. These engines then link to the shader cores, and the Cayman cards use a refined setup, known as VLIW4, which improves performance over previous generations. However, both Radeon HD 6970 and HD 6950 cards have fewer shaders than their year-old Radeon HD 5870/50 counterparts, which is rather surprising given the breakneck speed of GPU development.
AMD, though, attempts to mitigate the lack of shaders by boosting core and memory speeds. The table, below, compares the Radeon HD 6970/50 cards against established high-end competition.
| GPU | Radeon HD 5970 2,048MB | Radeon HD 6970 2,048MB | Radeon HD 6950 2,048MB | Radeon HD 6870 2,048MB | Radeon HD 5870 1,024MB |
| Transistors | 4.3bn | 2.64bn | 2.64bn | 1.7bn | 2.15bn |
| Die size | 2 x 334mm² | 389mm² | 389mm² | 255mm² | 334mm² |
| GPU | Cypress | Cayman | Cayman | Barts | Cypress |
| General clock | 725MHz | 880MHz | 800MHz | 900MHz | 850MHz |
| Shader clock | 725MHz | 880MHz | 800MHz | 900MHz | 850MHz |
| Memory clock | 4,000MHz | 5,500MHz | 5,000MHz | 4,200MHz | 4,800MHz |
| Memory interface | 256-bit x2, 2,048MB GDDR5 | 256-bit, 2,048MB GDDR5 | 256-bit, 2,048MB GDDR5 | 256-bit, 1,024MB GDDR5 | 256-bit, 1,024MB GDDR5 |
| Memory interface | 2 x 128GB/s | 176GB/s | 160GB/s | 134.4GB/s | 153.6GB/s |
| Price | £425 | £289 | £225 | £180 | £210 |
| GPU | NVIDIA GeForce GTX 580 1,536MB | NVIDIA GeForce GTX 570 1,280MB | NVIDIA GeForce GTX 480 1,536MB | NVIDIA GeForce GTX 470 1,280MB |
| Transistors | 3.0bn | 3.0bn | 3.0bn | 3.0bn |
| Die size | 520mm² | 520mm² | 529mm² | 529mm² |
| GPU | Fermi v2 | Fermi v2 | Fermi | Fermi |
| General clock | 772MHz | 732MHz | 700MHz | 607MHz |
| Shader clock | 1,544MHz | 1,464MHz | 1,401MHz | 1,215MHz |
| Memory clock | 4,008MHz | 3,800MHz | 3,696MHz | 3,206MHz |
| Memory interface | 384-bit, 1,536MB GDDR5 | 320-bit, 1,280MB GDDR5 | 384-bit, 1,536MB GDDR5 | 320-bit, 1,280MB GDDR5 |
| Memory interface | 192.4GB/s | 152GB/s | 177.4GB/s | 133.9GB/s |
| Price | £389 | £269 | £299 | £190 |
Architecture improvements
Cayman cards also include improvements to the areas of a GPU that are dedicated to image quality - antialiasing and anisotropic filtering - and both Radeon 6900-series GPUs are now equipped with 2GB of onboard RAM, to help smoothing out frame rates at high resolutions.
Cayman, too, is equipped with what AMD calls PowerTune technology. In a nutshell, hardware sensors on the cards determine the power required by the GPU and, should it exceed the thermal specifications when running a particular program, the GPU's frequency is turned down. AMD says that this will only happen in extreme cases, however, and it's ostensibly designed to safeguard the card and system in the case of a power surge instigated by a particular program.
Performance
But gamers want immediate frame-rate increases above all else. The £279 Radeon HD 6970 benchmarks at the same level as NVIDIA's GeForce GTX 570 card. This means it is not as fast as the current single-GPU champ, GeForce GTX 580.
Comparing Radeons, HD 6970 is on average around 20 per cent faster than the Radeon HD 5870 card released in September 2009. Meanwhile, the £225 Radeon HD 6950 is a touch quicker than the venerable HD 5870. The following chart illustrates how the high-end GPUs play out against one another in Just Cause 2, benchmarked at high-quality settings.
The numbers speak for themselves, though it's important to stress that AMD hasn't made the huge leaps in performance that are usually associated with new GPU architectures. Many observers had opined that the Radeon HD 6970 and HD 6950 GPUs would be faster than their NVIDIA GTX 580/570 competition, but improvements to the GeForce core via the 500-series architecture and AMD's transitional Cayman design combine to limit the comparative effectiveness of the two new Radeons.
Summing it up
If you're reading this TekSpek to gain a better understanding of how the Radeon HD 6970 and HD 6950 graphics cards work, please appreciate that AMD has changed the structure of its GPUs to better fit in with how future games will run on high-end cards. This means that, moving forward, they should be able to distance themselves from older Radeon cards such as the HD 5870.
There's now more focus on what's known as geometry processing, a more-elegant core setup, better image-quality filtering, and the PowerTune feature will be present on all future designs. Right now, though, these features make the Radeon HD 6970 and HD 6950 barely competitive against comparable NVIDIA cards. Take this back a year and AMD was comfortably ahead of price-comparable GeForce cards.
Sage buying advice would have you put the £225 Radeon HD 6950 2GB card on a shortlist. It offers an expansive feature-set and solid performance at high-quality settings. It also occupies a pricing position that NVIDIA is yet to exploit with its GTX 500-series Fermi GPUs.
