AMD Radeon HD 6970 and HD 6950 Review

After unexpected delays (and plenty of rumors that soon followed), AMD’s Radeon HD 6970 and Radeon HD 6950 finally see the light of day. The Cayman GPU brings a new VLIW4 core design that offers better efficiency and improved tessellation performance among others. AMD has positioned the HD 6970 as the superior enthusiast GPU, while the HD 6950 is touted to be in a class of its own. But still, the burning question: can the AMD Radeon HD 6970 subdue NVIDIA’s GeForce GTX 580?

With the new Radeon HD 6900 Series of GPUs, AMD set about achieving four primary design goals.

Efficient graphics and compute architecture

The first is to deliver a more efficient graphics and compute architecture. This is done by adopting a new VLIW4 core architecture with up to 24 SIMD engines and 96 texture units. On the VLIW4 design, all stream processing units now have equal processing capabilities and scheduling and register management has been simplified. This gives a 10% improvement in performance per mm squared when compared to the previous VLIW5 design, found on the Radeon HD 5000 Series. Render back-ends have been upgraded for improved anti-aliasing performance.

Massive geometry performance

Tessellation performance was poor with the Radeon HD 5000 Series. In the HD 6900 Series, AMD has implemented dual graphics engines. This gives double the transform and backface cull rate, tripled tessellation performance compared to the Radeon HD 5870 thanks to dual 8th generation tessellators and dual rasterizers for up to 32 pixels per clock.

New image quality features

Raw performance aside, the Radeon HD 6900 Series brings new image quality features including Enhanced Quality Anti-Aliasing (EQAA) and morphological anti-aliasing (MLAA).

EQAA offers enhanced quality over standard Multi-Sample Anti-Aliasing (MSAA) modes by doubling the number of coverage samples per pixel, while keeping the same number of color / depth / stencil samples. This technique offers advanced smoothing of aliased edges without requiring additional video memory, and with a minimal performance hit.

The MLAA technique works as a post process effect; each frame is first rendered normally and then run through another shader pass to perform the filtering before it goes to the display. This differs from traditional multi-sample and super-sample AA techniques where the filtering occurs during the rendering of each frame. AMD suggests that this technique can eliminate aliasing for still images, though it is intended to work better when in motion.

Power efficiency and power management

Lastly, AMD has also worked on power efficiency and power management. With PowerTune Technology, the GPU TDP is clamped to a pre-determined level. The GPU includes counters across all blocks which are monitored and applied to an algorithm to infer power draw. The core clock is then adjusted dynamically to enforce the TDP level.

For the Radeon HD 6900 Series, two GPUs are available: Radeon HD 6970 and Radeon HD 6950. The Radeon HD 6970 has a fully-enabled Cayman chipset with 24 SIMD engines for 1536 Stream Processors and 96 texture units. It is clocked at 880MHz core and 1375MHz (5500MHz effective) memory.

Definitely, not all silicon produced is good enough to meet the requirements to become a Radeon HD 6970 part. In a binning process, semi-defective silicon is screened again to find out whether they can be used in a slower part. The Radeon HD 6950 has two SIMD engines disabled, resulting in 22 SIMD engines and 88 texture units. The HD 6950 is also clocked lower at 800MHz and 1250MHz (5000MHz effective) for core and memory respectively.

Before we proceed, here is a quick specifications table.

With quite a number of noteable enhancements, how does the new AMD Radeon HD 6970 and Radeon HD 6950 fare? Let’s find out.