Evergreen (GPU family)

The Radeon HD 5000 series is a series of desktop graphics chips from AMD, which was introduced in the fall of 2009 as the successor to the Radeon HD 4000 series. All graphics processors of this series support Shader Model 5.0 DirectX 11 and OpenGL 4.1 after, also also OpenCL 1.2 in the current version. AMD marketed the ability of graphics cards other tasks away from the graphics computation to meet the term " ATI Stream". The successor to the Radeon HD 6000 series.

For the Radeon HD 5000 series AMD used the code name " Evergreen ". The mobile graphics solutions are marketed as ATI Mobility Radeon HD 5000 - series.

• 4.1 Performance
• 4.2 Power consumption data

• 5.1 Measurements of the power consumption

History

First executable models of the Evergreen series AMD presented in July 2009 at Computex in Taipei. On 10 September 2009, AMD before the Radeon HD 5000 series of professional press. The official launch of the Radeon HD 5850 and HD 5870 was held on 23 September 2009, but only test samples of the HD 5870 were supplied by the manufacturer. Test samples of the HD 5850 were delivered later, so that they could not be tested until 30 September 2009.

The Cypress - core, which is partly still referred to in the trade press as RV870, uses the so-called " Terascale 2 architecture" and was approximately 2.1 billion transistors at the time of conception until then most complex GPU on the market. The Cypress - core is fabricated in 40 - nm manufacturing process and thus a die area of 334 mm ². It is divided into multiple shader clusters, with a cluster of 16 5D shader units and four texture units. He thus takes over the distribution of RV770 and RV790 GPUs, however, the number of shader clusters of 10 to 20 increased compared to the previous generation has thus doubled the number of stream processors on the Cypress core of 800 to 1600. Independently of the shader clusters, the number of raster operation processors (ROP) is doubled. Since, however, continue a 256 -bit memory interface is used, two ROP clusters are now available per memory controller, which are surrounded by a 128 KB L2 cache.

The Radeon HD 5870 drew attention to the performance of the dual- graphics card Radeon HD 4870 X2 in her presentation about. Besides performance, AMD received in the trade press praise for the improved power saving mechanism after this limited effectively worked on the previous generation with the use of GDDR5 memory. On the Radeon HD 5850 two shader clusters disabled and also reduces the clock speeds. Although both cards the official presentation on 23 September were already listed, was not initially available, which is why it is this was a so-called Paper Launch. The availability of the cards initially improved either, for which purpose. Besides high demand, helped by a lack of competition from Nvidia products, primarily technical difficulties with the 40nm manufacturing process at contract manufacturer TSMC were causally Only when TSMC claims to be able to fix this in early 2010, a nationwide availability of the Radeon HD 5850 and HD 5870 was achieved.

On 13 October 2009, AMD finally introduced the Radeon HD 5750 and HD 5770, which are based on the Juniper chip. This is largely a halved version of the Cypress GPU, which has a die size of 166 mm ² with 1.04 billion transistors. In contrast to the Radeon HD 5870 and HD 5850, the HD 5770 and HD 5750 already coverage available on the market for the presentation were, the test samples failed again in the latter AMD timely deliver to the trade press. Performance-wise, AMD replaced with the card in about the same fast predecessors Radeon HD 4850 and HD 4870th

On 18 November 2009, AMD introduced the dual-GPU graphics card Radeon HD 5970th There are distinguished by two Cypress GPUs installed with a PCIe switch on the card. Graphics cards with two Cypress GPUs are led by AMD under the code name " Hemlock ". In order to comply with the official PCIe specifications and have a power consumption of less than 300 W, AMD reduced the clock speeds on the level of the Radeon HD 5850th On the deactivation of shader clusters could be dispensed with. At the launch of the Radeon HD 5970 this had a significantly higher performance than the hitherto fastest graphics card on the market, the GeForce GTX 295 Contrary to some expectations had AMD made ​​no improvements in the field of micro stuttering. A special feature of the Radeon HD 5970 is its length of 30.5 cm, which they do not in all mid-tower can be installed.

On 14 January 2010, AMD introduced the Radeon HD 5670, with which the Radeon HD 5000 series has been extended to the mainstream sector. The Radeon HD 5670 is based on the Redwood graphics processor, which is in the broadest sense is a halved Juniper chip. As the number of grid units was halved, but still the 128-bit memory interface is used, only one ROP partition is now again available for each memory controller. Performance-wise, the AMD Radeon HD 5670 placed between the Nvidia Geforce GT 240 competitors and Geforce 9800 GT.

On 4 February 2010, AMD released the Radeon HD 5450, the first low- end graphics card, the HD 5000 series. The HD 5450 is according to the reference design from AMD a passively cooled card in the Ultra Low Profile format, which is suitable for example for home theater PCs. The graphics card uses the Cedar graphics processor. In this, the structure of shader cluster is slightly different, it must be paired each with 8 shader units with 4 texture units, so Cedar of 2 clusters shader, but there is a total of 16 8 texture units. On 9 February 2010 was followed by the HD 5570, which is based on the same chip as the HD 5670, but clocked lower and is equipped with a slower memory. The same day presented Radeon HD 5550 is based on a teildeaktivierten Redwood GPU, but was due to the low support of the board partners do not market relevance. A little later, on February 25, completed for the time being, the Radeon HD 5830 product family. The model is placed in price between the Radeon HD 5770 and HD 5850. The graphics card uses the major RV870 chipset, which has been severely limited, so that now only 1120 stream processors and 56 TMUs 16 ROPs of the Radeon HD 5830 are available. But compared to the Radeon HD 5850 clock speeds were slightly increased, but this results in a slight increase in power consumption.

Technology

Architecture

With the Radeon HD 5000 series AMD used the first time the so-called " Terascale 2 architecture ", where it is an evolution of the unified shader architecture of the R600 GPU. Due to the design for DirectX 11 different changes had to be made, including the support of Shader Model 5.0. This includes the use of the new compute, Hull and Domain shaders (6:1 compression, for example, for 16- bit HDR textures) made ​​the improvement of texture compression is necessary and required by the texture units the maximum texture resolution 16k x 16k (previously 8k ). The compute shader is support for DirectCompute, ie the calculation of non-graphical data on the GPU ( GPU computing ). DirectCompute is competing to OpenCL. The architecture is optimized for GPU computing and supports the IEEE - 754-2008 standard, which is why the GPU Fused Multiply Add ( FMA) can perform. In addition to improvements to the caches were still further commands, such as Sum of Absolute Differences (SAD) as well as important for DirectCompute commands such as Bit count, insert, extract, implemented. Since the R600 graphics processor installed ATI called tessellator units, which are generally necessary with DirectX 11 support. With them it is possible to relatively low computational effort to increase the number of polygons in an object. What is new is the link to the Hull and Domain shaders, while largely acted on the previous generation of tessellator only with the geometry shader. Regardless of the DirectX 11 implementation presents AMD under the name " ATI Eyefinity " the ability to control more than two monitors natively.

The " Terascale 2 architecture" uses a new memory management, which now dominates EDC ( Error Detecting Code ), the error due to cyclic redundancy check ( CRC) are detected. This is primarily relevant in the field of GPU computing tasks, which is treated with AMD, the term " ATI Stream". Through so-called "Link Retraining ", it is also possible to change clock and voltage of the memory fast and reliable, which is why the GDDR5 memory is in contrast to the previous series automatically clocked down deep even at idle.

GPUs

Naming

The Evergreen series the same designation system as they have for the Radeon HD 3000 series control panels. All graphics cards are referred to as " ATI Radeon HD" and an additional four-digit number that generally begins with a "5" ( for the series ). The second and third digits are the subdivision into different models.

Distribution:

• HD 5450 and 5550: low-end
• HD 5570 and 5670: Mainstream
• HD 57xx and 5830: Performance
• HD 5850 /70 and 5970: High- End

Given the general slump in market prices and currency fluctuations, the original classifications of ATI do not apply generally.

Model data

800 (1600)

GDDR5

900 (1800)

GDDR5

Performance

For each model, the following theoretical power data:

Notes:

• The above performance data for the computing power of the stream processors, the pixel fill rate, the Texelfüllrate and memory bandwidth are the theoretical maximum values. The overall performance of a graphics card depends, among other things, on how well the available resources can be exploited or utilized. In addition, there are other not listed here, factors that affect the performance.
• The computing power of the stream processors is not directly comparable with the performance of Nvidia Geforce series, since this is based on a different architecture, which scales differently.

Power consumption data

The measured values ​​listed in the table refer to the pure power of graphics cards that comply with the ATI reference design. In order to measure these values ​​, it requires a special measuring apparatus; depending on the employed measurement technique and given measurement conditions, including the shared program, with which the 3D load is generated, the values ​​between different equipment may vary. Therefore, here measured value ranges are specified, each representing the lowest and highest measured values ​​from different sources.

Far more common than the measurement of the consumption of the graphics card is to determine the power consumption of a whole system. To this end, a reference system is assembled, in which the various graphics cards to be installed; then finds the measurement using an energy cost meter or similar apparatus rather than directly to the outlet. However, the significance of the measurements is limited: it is not clear which consumption comes from the graphics card and what is attributable to the rest of the PC system. The difference in consumption between idle and load 3D operation of this measuring method is not only dependent on the program with which the load was produced; the utilization and efficiency of the rest of the PC system, including the power supply, motherboard and processor affect the measured difference also. Because different the tested systems are usually of one's own home PC system, the values ​​given there can not be mapped to its own system. Only data from otherwise identical systems are good (conditional) for comparison with each other. Because of this dependence overall system measurement values ​​are not listed in the local table. But since they can provide a better picture of the practical power of a concrete system with a certain graphics card, web pages are listed under the links that are undertaking such measurements.