Radeon HD 7000 Series

The Radeon HD 7000 series is a series of desktop graphics chips from AMD and successor to the Radeon HD 6000 series. With the Radeon HD 7000 series, which also bears the code name "Southern Iceland ", AMD introduced the first one to support DirectX 11.1. All graphics processors of this series support Shader Model 5.0 DirectX 11 and OpenGL 4.1 after, cards with the nuclei Cape Verde, Pitcairn & Tahiti also support DirectX 11.1 and OpenGL 4.2 and PCI Express 3.0. The interfaces DirectCompute 11.1, OpenCL 1.2 and C AMP can be used with the DirectX 11.1 -capable models for universal calculations.

  • 2.1 GPUs
  • 2.2 Model data
  • 2.3 Performance data recording

Description

History

The first graphics card in the Radeon HD 7000 series, AMD ago on 22 December 2011. This was the Radeon HD 7970, which was based first on the Tahiti GPU. This, internally referred to as R1000, presented in more ways than one oddity: He was the first graphics chip, which was created in 28- nm manufacturing process, which allows the use of 4.31 billion transistors enabled. This set of R1000 end of 2011, until then the most complex GPU on the market dar. Since the Radeon HD 6900 card uses the AMD " PowerTune " mentioned technique for limiting the maximum power consumption by the card. The same technique is now also in the Radeon HD 7000 series to use. Due to this limitation are achieved in practice maximum power values ​​, such as GFLOPs the GPU, even further away from the theoretical values ​​than was the case for graphics cards without this capping of maximum power.

The official launch of the Radeon HD 7970 was held on 9 January 2012. The card had a higher by about 20% performance compared to the GTX 580 (up to 35 % in extremely high resolutions ) and placed so that initially the fastest single - GPU card on the market dar. in the trade press, the energy efficiency of the card was positively rated, partly because AMD using the "Zero Core power" feature presented an improved power saving mode. The criticized at the Radeon HD 6900 cards anisotropic texture filtering revised the AMD R1000 completely, but you gave up the implementation of new bildqualitätsverbessernder features. As the biggest criticism of the Radeon HD 7970 proved to be the reference cooler, reaching under 3D applications over 4 sone.

January 31, 2012 AMD introduced the Radeon HD 7950 before the second graphics card based on the R1000 GPU. In contrast to the Radeon HD 7970 the presentation day was also the launch. Four of the 32 shader clusters of Tahiti GPU were teildeaktiviert on the Radeon HD 7950, making this one about 5 % higher performance compared to the preliminary Nvidia competitor, the GeForce GTX 580 reached. As with the Radeon HD 7970 AMD raised the official selling price compared to the previous series: While the starting price, the Radeon HD 6950 was still at 260 €, AMD gave these to the Radeon HD 7950 420 €.

On March 5, 2012, AMD introduced the Radeon HD 7850 and 7870 cards, which replaced the Radeon HD 6800 series. The maps are based on the Pitcairn GPU that comes with 2.8 billion transistors to 1280 shader processors and 80 texture units in 20 clusters. On the Radeon HD 7850, four of these clusters are disabled. Compared to their predecessors recognized the cards on a performance jump of up to 40%, which was positively evaluated in combination with the lower power consumption, due to the 28- nm process in the trade press. However, AMD also highlighted in these models the list price ( U.S. $ 249 for the Radeon HD 7850, or U.S. $ 349 for the Radeon HD 7870 ) and are thus initially have a lower price - performance ratio. Official launch was planned for 19 March 2012.

On January 4, 2012, AMD Radeon HD 7350 graphics cards, 7450, 7470, 7570 and 7670 before. This is not really concerns however about "new" cards, but actually for the OEM market renamed graphics cards from the AMD Radeon HD 6000 series ( " rebranding " ), which have only marginal changes.

Architecture

The technical basis of the Southern Iceland series is the newly developed GCN architecture ( "Graphics Core Next " ), which replaced the previous VLIW architecture. The primary advantage of the new instruction architecture compared to the previous VLIW architecture is that dependencies of the instructions do not result in the code to have empty some ALUs. In the previous VLIW architecture four instructions were distributed to 16 VLIW shader ever. Goods within these four statements some interdependent, had the dependent instructions to wait until the dependencies have been resolved by appropriate ALUs, the VLIW shaders remained unused. The SIMD ALUs, the current architecture are summarized respectively in 16s groups ( Vec16 ). However, each Vec16 SIMD block can execute a statement independent of other SIMD blocks, making the utilization of ALUs are independent of dependencies of the instructions. It can be carried out with a statement within a SIMD Vec16 unit at 16 different input data streams similar calculation Opara TION. The basic structure of the SIMD units is also similar to the VPU of Larrabee, there also exist the VPU from Vec16 SIMD units. Underutilized ALUs can occur but also in the new architecture, but with a completely different scenario: If a statement to less than 16 data streams must be applied and thus the Vec16 SIMD block is not completely filled. The biggest advantage of the new architecture is that the programming is easier, because the compiler now does not have to try the instructions to pack so that they can be distributed to the VLIW shader with as few interdependencies. Accordingly, all other programming tools and the flexibility to simplify, also increases with respect to GPGPU applications.

A shader clusters, even Compute Unit (Eng. Processing Unit) called, consists of four Vec16 SIMD units, each SIMD vector unit has its own register with the overall size of 64 KB available. Each shader clusters also have an additional 64 KB of local cache ( "Local Data Store " ) are available, can share access to the all SIMD blocks. Moreover, a shader cluster has also an independent scalar unit for data management ( address calculations, management of the data flow, etc.) and other, simple, scalar calculations is responsible within the shader cluster. A scheduler distributes the instructions on the four Vec16 SIMD blocks within the shader cluster. outside the shader cluster and it is necessary 16 KB large L1 caches available, which can be accessed on the respective texture units and to each memory controller are another 128 KB cache connected.

Naming

In the current series, the same designation system as well as its predecessor Radeon HD 6000 series control panels. All graphics cards are referred to as "AMD Radeon HD" and an additional four-digit number that generally begins with a "7" ( for the series ). The second and third digits are the subdivision into different models. The mobile graphics chips to go under the model designation AMD Radeon M.

Distribution:

  • HD 7350 to HD 7670: low-end graphics cards and for the OEM market
  • HD 77xx: Mainstream
  • HD 78xx: Performance
  • HD 79xx: High- End

Data Overview

GPUs

Model data

Power consumption data

Comments

55698
de