ASUS EAH6970 2GB Review

👤by Sahil Mannick Comments 📅27-12-10

Cayman is the rightful heir to the successful Cypress core. For the most part, the architecture is much the same with some tweaks to increase efficiency and performance per die size. The GPU comes in two variants, Cayman Pro and Cayman XT, the former found on board the HD6950 and the latter on the HD6970. Originally, the Cayman core was poised to be manufactured on the 32nm manufacturing process which would have brought along many new features and possibly better performance in the same package. However, TSMC cancelled the process so Cayman had to resort to the more mature 40nm process, sacrificing many of the new features promised.

The biggest change to the architecture is the use of the VLIW4 design instead of the VLIW5 design used since DX9 graphics cards were first introduced. The main reason was to increase efficiency in newer DX10/11 titles and to put more emphasis on GPGPU computing. The VLIW4 architecture introduces a narrower Stream Processing Unit, in which the 5th stream processor named the “t-unit” has been removed leaving just 4. Despite the Unit only being able to process 4 operations instead of 5, the advantage is that the space saved allows the core to have more SIMD clusters. The end result is that Cayman can have up to 24 SIMD clusters (24 on the HD6970 and 22 on the HD6950) whereas Cypress only had up to 20 SIMDs (20 on the HD5870 and 18 on the HD5850), giving Cayman a 10% improvement per mm squared compared to previous VLIW5 architecture (as used on Barts also). The architecture is now more catered towards GPU Compute than before and the higher SIMD count also gives additional texturing performance. Each SIMD from both architectures contains 4 texture units so Cayman now benefits from up to 96 units rather than the 80 available on Cypress.

Cayman: VLIW4, 24 SIMD clusters, 96 texture units
Cypress: VLIW5, 20 SIMD clusters, 80 texture units

Another benefit of the new core is its improved dual 8th generation tessellator that bring up to 3 times the tessellation performance of the HD5870.

As mentioned, Cayman comes in two variants as found on the HD6950 and HD6970. The key difference is that the HD6950 has 22 SIMD clusters instead of 24 on the HD6970, leading to fewer texture units (88 vs. 96) and fewer stream processors (1408 vs. 1536). The GPU boasts an impressive 2.64 billion transistors on a die size 389mm squared. For reference, the mainstream Barts core sports 1.7 billion transistors on a 255mm squared package. A 256-bit GDDR5 memory bus is still supported but this time 2GB of VRAM is provided to compete with Nvidia’s offerings.

Cayman is a step ahead of Cypress and Nvidia when it comes to texturing performance but the stream processor deficit means that the pixel fill rate and shader operation can’t quite match up to the HD5870. The question is whether the more efficient design can give better performance in real gaming situations. One downside of the Cayman core is its maximum power consumption but that is always the risk with packing more transistors into ever larger cores. AMD have PowerTune to regulate the TDP so we don’t expect the power draw to exceed 250W. More will be revealed about PowerTune later on.

Product Specifications

Graphics Engine: AMD HD6970
Model: ASUS EAH6970 2DI2S/2GD5
Bus Standard: PCI Express x16 2.1 (Compatible with 1.1)
Memory Size (MB): 2048 GDDR5
Memory Interface: 256-bit
Core Clock Speed (MHz): 890
Stream Processors: 1536
Shader Clock (MHz): 890
Memory Clock Speed (MHz): 5500

3D API: Direct X 11 & OpenGL 4.0
DVI Output: 2 (DVI-I & 1 DVI-D)
HDMI: 1 (HDMI 1.4a w/audio (8 channel LPCM & bitstream))
DisplayPort: 2 mini DPs 1.2 (1 DP-to-mini DP adapter supplied)
VGA: Supported using DVI-to-VGA adapter
Display Output (Max Resolution): 2560 x 1600
Crossfire Support: Yes
Card Dimension (mm): 274 x 115 x 35

25 pages 1 2 3 4 > »