Page 2 - AMD Radeon HD 7800 Series Architecture
AMD's Southern Islands family of GPUs is built on the 28nm fabrication process, and are PCIe 3.0 compatible cards. Separated into three market segments, each chip series carries its own codename -- the entry level Radeon HD 7700 is called Cape Verde, while the high end Radeon HD 7900 is coded Tahiti. The mainstream performance Radeon 7800 series card we are covering today is named Pitcairn. The Gigabyte Radeon HD 7870 2GB OC we are examining today is a non-reference board that comes with the company's own WindForce 3X cooler in conjunction with a factory overclock for a price of approximately $330 at press time. This is the average price for a Radeon HD 7870, and you will be happy to know it comes with a 3-year warranty, factory overclock, and an excellent custom cooler to boot. We will talk about the performance and noise of it later on in this review.
This DirectX 11.1 card from AMD has a default core clock of 1100MHz, and 256-bit memory running at an effective data rate of 4800MHz. This makes our Gigabyte Radeon HD 7870 2GB OC is 100MHz faster than stock, but memory is kept at stock according AMD's specifications. Compared to the 6-series video cards, AMD has moved away from the VLIW5 architecture for improved computing performance, while retaining graphics rendering performance and lowered power consumption. The new Graphics Core Next (GCN) architecture is essentially one shader cluster for each compute unit; consisting of four SIMD-16 vector units, four texture filter units, and sixteen texture fetch load/store units. Each GCN compute unit has its own vector registers, scalar registers, local data share, and L1 cache. Because the GCN architecture no longer require instructional level parallelism, it is much more optimized for processor utilization; in other words, performance in both gaming and computing as aforementioned.
Putting this all together, AMD managed fit 2.8 billion transistors on the chip. This is quite a number up from the previous generation Barts CPU, which consists of 1.7 billion transistors. The Radeon HD 7870 we are reviewing today has 1280 shader processors in aggregate spread across 20 compute units, with 32 color ROPs per clock cycle, and tied to 512KB of L2 cache. (Just for reference, the Radeon HD 7850 has 16 compute units.) This brings you 2.56 teraflops of processing power at a reduced die size from 255mm^2 to 212mm^2, combined with a full two gigabyte of 256-bit GDDR5 memory. AMD's graphics processor retains an excellent 10W TDP at idle, down from 19W with the 6870, and maxes out at 175W rated TDP under full throttle.
As far as technical improvement highlights over the previous generation, the first thing I would like to talk about is DirectX 11.1 support. DirectX 11.1 will come with Windows 8, bringing support for things such as WDDM 1.2, as well as better Direct2D, Direct3D, and DirectCompute integration. Secondly, the Radeon HD 7870 has even better tessellation performance. The Tesselator Gen 9 claims to improve numbers up to four times compared to Gen 8, with generally improved numbers across the board. Eyefinity 2.0 will let you configure custom resolutions with different sized monitors at up to 16000x16000 total resolution. The new Taskbar Positioning feature will let you set the monitor you want your taskbar to be on. AMD's Radeon HD 7870 also has better Stereo 3D support over HDMI 1.4a. The Unified Video Decoder (UVD) is updated with hardware DivX and MPEG-4 support. And last but not least, AMD's Dual Stream HD+HD allows you to stream two 1080p videos at once. Not sure why you would need that, but it is nice to have, haha.
Page Index
1. Introduction, Specifications, Bundle
2. AMD Radeon HD 7800 Series Architecture
3. A Closer Look, Test System
4. Benchmark: 3DMark 11
5. Benchmark: Battlefield: Bad Company 2
6. Benchmark: Call of Duty: Black Ops
7. Benchmark: Colin McRae: DiRT 2
8. Benchmark: Far Cry 2
9. Benchmark: Just Cause 2
10. Benchmark: Metro 2033
11. Benchmark: Unigine: Heaven v2.5
12. Power Usage, Temperature, Noise
13. Overclocking and Conclusion
