ATI Radeon HD 2000 series

The Radeon HD 2000 series is a series of desktop graphics chips from AMD and the successor to the Radeon X1000 series . It is the sixth generation of graphics processors with the name ATI Radeon . All graphics processors in this series support pixel , geometry and vertex shaders 4.0 according to DirectX 10. The mobile graphics solutions are marketed as the ATI Mobility Radeon HD 2000 series . It was replaced by the ATI Radeon HD 3000 series .

history

Radeon HD 2400 XT from Sapphire with passive cooling

The architecture of the R600 graphics processors and their derivatives, which are used in the Radeon HD 2000 series, is not completely new , although it has never been used in graphics cards for PCs . It is based in principle on the Xenos architecture of the graphics processors of the Xbox 360 , which were codenamed R400. At the time the architecture for the Xbox 360 was created, it was planned to also create the desktop version on this architecture. This idea was then rejected, however, and both the Radeon X series and the Radeon X1000 series were still based on the second Radeon architecture, which was used from the Radeon 8000 series .

According to AMD's initial plans, the Radeon HD 2000 series should be released before Christmas 2006. However, this date could not be kept, in fact it was introduced both after Windows Vista (with which DirectX 10 was introduced) and after Nvidia's competitor series GeForce 8 . Nvidia even managed to present not only the high-end models before AMD, but also the mainstream cards were presented and sold in April 2007, and thus before the appearance of the HD-2000 series.

In contrast to the competition, AMD not only presented the high-end model (HD 2900 XT) on May 14, 2007, but also presented the technical data of the mainstream graphics cards (HD 2600) and the low-end graphics cards (HD 2400). However, only the Radeon HD 2900 XT was immediately available for sale.

Surprisingly, AMD put the top model at the launch of the HD 2000 series, the Radeon HD 2900 XT, not against the top models Nvidia, the GeForce 8800 Ultra or GTX, but only against the somewhat slower 8800 GTS in terms of price and performance. In this regard, the Radeon HD 2900 XT could not convince in all respects. In addition, the Radeon HD 2900 XT has a very high power consumption, which exceeds all - even faster - desktop graphics cards on the market at the time of its introduction. The Catalyst 7.8 driver released in August 2007 brought a little more performance under AA and AF . The Catalyst 7.10 (October 2007) fixed further driver errors (although not all) under Windows Vista and increased the performance of Crossfire systems . A "miracle driver" hoped for or expected by some that would bring the performance of the HD 2900 XT to a level that corresponds to the high power consumption did not materialize.

On 28 June 2007 was followed by the complete presentation of the mainstream and low-end graphics cards, which, among other things, in the case of disclosure agreements (NDAs) for benchmarks reflected. The first graphics processors produced using 65 nm manufacturing technology were exactly the opposite of the Radeon HD 2900 XT in terms of power consumption: The HD-2400 series set new records for DirectX 10 graphics cards in terms of low power consumption, but so did the HD-2600 Series was very frugal. In the graphics processors used in these series, the Unified Video Decoder is also installed for the first time , with which film material can be decoded via the graphics cards without a large processor load.

In autumn 2007 further additions to the product range followed. First, the Radeon HD 2600 XT X2 was presented by the board partners in September. A completely new circuit board layout was developed for this graphics card, as it has two graphics processors on one graphics card. This was also reflected in the sales price, which is why relatively few batches were produced. Since AMD placed a high-end dual graphics chip card with the Radeon HD 3870 X2 at the beginning of 2008, the experience gained through the development should also have been an argument for producing it.

As the successor to the R600 was due at the end of 2007 with the RV670 (HD-3800 series), AMD introduced the Radeon HD 2900 Pro and the Radeon HD 2900 GT in autumn 2007, which were to be used to sell the stocks of R600 chips. They were based on the same circuit board layout as the Radeon HD 2900 XT and were also equipped with the same cooler design. Due to the placement as a product to sell out the R600, both graphics cards were very inexpensive and only briefly on the market.

Technical

Unified shaders

As with Nvidia's competing series, the Radeon HD 2000 series uses unified shaders for the first time. The evolution of technology has led to the fact that one can no longer really differentiate between quads in the conventional sense, since there are no longer any rendering pipelines , but compound units still exist. The tasks of the pixel pipelines, vertex and pixel shaders from the rendering pipelines are now also handled by so-called stream processors (SPs).

However, these differ significantly from the competition: Each five stream processors are combined to form a shader unit. Depending on the model, four to 16 of these quintets form a shader cluster, which consequently consists of 20, 40 or 80 stream processors. Within a shader unit, all stream processors can perform multiplication and addition operations (MAD), but one also has the ability to perform transcendent functions (e.g. sine and logarithmic calculations). These five stream processors can independently execute scalar instructions, what AMD calls "superscalar". At the same time, however, they also work in parallel, with the result that no stream processor can compute an operation that is dependent on the result of an operation of another stream processor in this shader unit. This can result in individual stream processors running empty, provided the driver cannot find any other useful calculations and thus the theoretical maximum performance of the graphics processor cannot be used.

The texture units (TMUs) are arranged parallel to the shader clusters and form a TMU cluster. The number of TMUs corresponds to the size of the shader cluster. This means that the GPU with small shader clusters (RV610) has 4 TMUs, the one with medium cluster size 8 TMUs and the R600 finally 16 TMUs. In the case of the R600 architecture, this enables the shader / TMU ratio to be selected very finely and texture and shader performance to be designed independently of one another within certain limits.

Graphics processors

Different graphics processors are used within the Radeon HD 2000 series, which differ in terms of their 3D capabilities.

Graphics chip	production			units						DirectX / OpenGL version	Video processor	cutting stelle
	process	transis- interfere	The - area	ROPs	Unified shaders			Texture units
					Stream processors	Shader units	Shader cluster	TAUs	TMUs
RV610	65 nm	180 million	082 mm²	04th	040	08 × 5D- VLIW	2	08th	04th	10.0 / 3.3	UVD 1.0	PCIe
RV630	65 nm	390 million	153 mm²	04th	120	24 × 5D-VLIW	3	16	08th		UVD 1.0
R600	80 nm	720 million	408 mm²	16	320	64 × 5D-VLIW	4th	32	16		-

Naming

All graphics cards are identified with a four-digit number that generally begins with "HD 2". The second digit then divides the family into different market segments. The third and fourth digits serve for further diversification. These models are further divided into different variants. These are marked with a corresponding abbreviation after the model number.

division

• HD 29xx: performance and high-end
• HD 26xx: mainstream
• HD 24xx: low-end

Letter abbreviations

• GT - budget version of a chip with stream processors disabled
• Pro - standard version of a chip, compared to the XT lower clock rates and sometimes slower memory
• XT - powerful version in all segments

Model data

model	Official launch	Graphics processor (GPU)							Graphics memory				TDP ( watt )
		Type	Active units					Clock rate (MHz)	Size ( MB )	Clock rate (MHz)	Type	Storage interface
			ROPs	Shader - cluster	Stream processors	TAUs	TMUs
Radeon HD 2350	Jun 28, 2007	RV610	4th	2	40	8th	4th	525	128 256	400	DDR2	64 bit	25th
Radeon HD 2400 Pro		RV610	4th	2	40	8th	4th	525	128 256	400	DDR2	64 bit	25th
Radeon HD 2400 XT		RV610	4th	2	40	8th	4th	700	256	700	GDDR3	64 bit	25th
Radeon HD 2600 Pro		RV630	4th	3	120	16	8th	600	256 512	400	DDR2	128 bit	45
Radeon HD 2600 XT		RV630	4th	3	120	16	8th	800	256 512	700	GDDR3	128 bit	50
										1100	GDDR4
Radeon HD 2600 XT X2	26 Sep 2007	2 × RV630	2 × 4	2 × 3	2 × 120	2 × 16	2 × 8	800	2 × 256 2 × 512	700	GDDR3	2 × 128 bits	100
Radeon HD 2900 GT	Nov 6, 2007	R600	8th	3	240	32	16	600	256	800	GDDR3	256 bit	150
Radeon HD 2900 Pro	25 Sep 2007	R600	16	4th	320	32	16	600	512	800	GDDR3	512 bits	150
									1024	925	GDDR4
Radeon HD 2900 XT	May 14, 2007	R600	16	4th	320	32	16	742	512	828	GDDR3	512 bits	215
									1024	1000	GDDR4

Hints

• The specified clock rates are those recommended or specified by AMD. However, the final specification of the clock rates is in the hands of the respective graphics card manufacturer. It is therefore entirely possible that there are or will be graphics card models that have different clock rates.
• Information in round brackets means that there are common retail versions from various manufacturers with this equipment, but that these are not listed as reference designs in public AMD documents.
• The date indicated is the date of the public presentation, not the date of availability of the models.
• The manufacturer's TDP information may differ from the real maximum power consumption.

Performance data

The following theoretical performance data result for the respective models:

model	Graphics processor / memory
	Computing power of the stream processors in GFlops	Pixel fill rate in GPixel / s	Texel fill rate in GTexel / s	Data transfer rate in GB / s
Radeon HD 2350	042	02.1	02.1	006.4
Radeon HD 2400 Pro	042	02.1	02.1	006.4
Radeon HD 2400 XT	056	02.8	02.8	011.2
Radeon HD 2600 Pro	144	02.4	04.8	012.8
Radeon HD 2600 XT	192	03.2	06.4	022.4
				035.2
Radeon HD 2900 GT	288	07.2	07.2	051.2
Radeon HD 2900 Pro	384	09.6	09.6	102.4
				118.4
Radeon HD 2900 XT	474.88	11.8	11.8	105.98
				128

Hints

• The specified performance values ​​for the computing power via the stream processors, the pixel fill rate, the texel fill rate and the memory bandwidth are theoretical maximum values. The overall performance of a graphics card depends, among other things, on how well the available resources can be used or fully utilized. There are also other factors that are not listed here that affect performance.
• The computing power via the stream processors is not directly comparable with the performance of the Nvidia Geforce 8 series, as it is based on a different architecture that scales differently.

Individual evidence

1. ↑ Computerbase: ATi Radeon HD 2900 XT - Power consumption under Windows XP , test report from May 14, 2007, accessed on February 3, 2010
2. ↑ Computerbase: ATi Catalyst 7.x - Driver comparison under Windows Vista , test report from October 13, 2007, accessed on February 3, 2010
3. ↑ Computerbase: ATi Radeon HD 2400 XT, HD 2600 XT, HD 2600 XT X2 and HD 2900 XT power consumption under Windows Vista , test report from September 26, 2007, accessed on February 3, 2010
4. ↑ Heise: dual graphics card from AMD , message from November 20, 2007, accessed on February 3, 2010

Web links

• Official product page
• ATI Radeon HD 2900 XT Review (www.3dcenter.de) (including considerations about the new architecture)