Technical data / specifications

In December last year, AMD paperlaunched their new Radeon HD 7970 which was developed under the codename Southern Island. The two biggest advancements concern their new Graphics Core Next architecture and the memory interface which is now 384 bit wide as well as the jump in process manufacturing technology from 40 nanometer to 28 nanometer. Actually that was quite a lot of work, especially if you keep in mind, that this chip, with its 4.31 billion transistors, has an insane complexity.

With the new architecture AMD did not only improve performance, they also optimized the anisotropic filtering and they added super sampling anti aliasing for DirectX 10 as well as DirectX 11, which helps improving overall display quality in games or generally 3D applications. Lets stay with the improvements. Meanwhile AMD also offers broader support GPU computing and hopefully they'll, at least at some point, be able to keep up with NVIDIA. AMD has also improved power consumption when the card is in idle mode. The ZeroCore power feature almost completely deactivates the GPU when the display(s) have been turned off.

The Radeon HD 7950 itself is based on the exact same GPU like the Radeon HD 7970. It also counts 4.31 billion transistors and it being manufactured using TSMCs latest 28 nanometer process technology. The differences can be found as soon as we dig a bit deeper. The Radeon HD 7950 comes with a total of 28 CUs whereas the Radeon HD 7970 has 32. As a result you get 1'792 scalar ALUs which are able to perform one MADD (Multiply Add) operation per clock. Furthermore every CU comes with four texture units and additionally there are 16 Load/Store-Units which do texture fetch operations. Finally you end up with 112 TMUs and 448 L/S-Units.

Regarding clock speeds the Radeon HD 7950 reference design posts 800 MHz for the GPU and 2'500 MHz (effective 5'000 MHz) for the memory. There is a total of 3'072 Megabyte of GDDR5 graphics memory and this gets the data via a 384 bit wide interface. Finally we end up with 240 Gigabyte per second of raw memory bandwidth. Thanks to the new 28 nanometer manufacturing process technology it has also been possible to keep the thermal design power at a reasonable level of 200 Watts.

Of course there are some more features. Therefore you can find PCI Express 3.0 support, DirectX11, UVD 3.0 and VCE (Video Codec Engine).

	Radeon HD 7970	ASUS HD 7950 DirectCUII Top	Radeon HD 6970	Radeon HD 6950
Chip	Tahiti	Tahiti	Cayman XT	Cayman Pro
Process	28 nm	28 nm	40 nm	40 nm
Transistors	4.31 billion	4.31 billion	2.64 billion	2.64 billion
GPU clock	925 MHz	800 MHz	880 MHz	800 MHz
Shader clock	925 MHz	800 MHz	880 MHz	80 MHz
Memory	3'072 MB GDDR5	3'072 MB GDDR5	2'048 MB GDDR5	2'048 MB GDDR5
Memory clock	2'750 MHz	2'500 MHz	1'375 MHz	1'250 MHz
Memory interface	384 Bit	384 Bit	256 Bit	256 Bit
Memory bandwidth	264'000 MB/s	240'000 MB/s	176'000 MB/s	160'000 MB/s
TMUs	128	112	96	88
TAUs	128	112	96	88
Shader Cores	2'048 (1D)	1'792 (1D)	384 (4D)	352 (4D)
ROPs	32 ROP	32 ROP	32 ROP	32 ROP
Shader model	SM 5.1	SM 5.1	SM 5	SM 5
Maximum board power	250 Watt	200 Watt	250 Watt	200 Watt

As you might expect the ASUS HD 7950 DirectCU II Top is everything but standard. First of all core clock speeds have been increased from 800 MHz to 900 MHz. For factory overclocked cards ASUS uses "Top" in the name. Furthermore ASUS equipped this card with its own DirectCU II cooler. This massive triple slot cooler features two 100 millimeter fans as well as five heatpipes. Each of which has a diameter of eight millimeters. In 2D mode card is almost inaudible and under load it still is much quieter than the AMD reference design.

Page 1 - Introduction	Page 8 - Stone Giant
Page 2 - Technical Data / Specifications	Page 9 - World in Conflict
Page 3 - Preview / Delivery	Page 10 - Resident Evil 5
Page 4 - Testing conditions	Page 11 - Call of Juarez
Page 5 - 3D Mark 11	Page 12 - Far Cry 2
Page 6 - 3D Mark Vantage	Page 13 - Overclocking
Page 7 - Unigine Heaven	Page 14 - Conclusion

Discuss this article in the forums