Architecturally, Cayman features a few more upgrades from Barts. One of the key changes in Cayman is that the chip now boasts dual graphics engines for more optimized usage and execution of graphics processing. On top of that, it also has a further improved tessellation unit, upgraded render back-ends, and a new VLIW4 core design. Overall, these optimizations lets the new Radeon HD 6900 do more per clock, which improves overall performance per core, and also drastically improves tessellation and anti-aliasing performance. In fact, AMD claims that the new the Radeon HD 6970 has three times the tessellation capability of the old Radeon HD 5870. This is a much welcome improvement seeing that even AMD’s latest Radeon HD 6870 and HD 6850 have mostly trailed NVIDIA’s Fermi cards when it came to tessellation-intensive application.
One of the new features of the Cayman cards is AMD's PowerTune Technology, which helps the card achieve better performance by dynamically adjusting the card's clock speeds during run-time based on an internally calculated GPU power assessment. Essentially, this means that instead of limiting the card's clock speeds to ensure that it operate within its TDP envelop, the Cayman cards are clever enough to adjust and even overclock the card if it sees that it is still operating well within its TDP envelop. Unfortunately, we couldn't give this option a go as PowerTune option did not show up on our version of Catalyst. As such, we'll probably delve into this topic on another occasion to highlight the extent of its merits.
The Radeon HD 6900 series also introduces to us EQAA, a new anti-aliasing option. It improves graphics quality by doubling the number of coverage sample per pixel while keeping the number of color, depth, and stencil samples constant. This offers smoothing of aliased edges with little penalty to video memory and at a minimal performance cost. This can be enabled using AMD's Catalyst software.
If you are interested in numbers, the Radeon HD 6970 has 24 SIMD engines, 1536 stream processing units and runs at a considerable 880MHz at the core. That’s slightly lesser than the 1600 stream processing units that the old Radeon HD 5870 has, but then again, the cores are now more efficient than before, so it is unfair to compare them outright. To add, the Radeon HD 6970 also has a generous 2GB of framebuffer which runs at staggering 5500MHz GDDR5. Memory interface continues to be limited to 256-bits wide, but thanks to the ultra-high memory clock speeds, effective memory bandwidth is now a whopping 176GB/s.
Not to be outdone, the Radeon HD 6950 has 22 SIMD engines and 1408 stream processing units - lesser than the Radeon HD 6970 but still beefy nonetheless. The Radeon HD 6950 also has a large 2GB framebuffer with a 256-bit wide memory interface, but it will run at slightly lower clock speeds - 800MHz at the core and 5000MHz DDR at the memory, to be exact.
Here are some photos of the cards.