AMD Radeon HD 6000 series

Radeon HD 6870 from Sapphire

The Radeon HD 6000 series is a series of desktop - graphics chip company AMD , which in autumn 2010 as a successor to the Radeon HD 5000 series was introduced. All graphics processors in this series support the Shader model 5.0 according to DirectX 11 and OpenGL 4.4. The Radeon HD 6000 series, which also bears the code name "Northern Island", is for the first time no longer marketed under the ATI logo, but directly under AMD's logo. The successor is called “ AMD Radeon HD 7000 Series ” and came onto the market in spring 2012.

description

The first models of the Radeon HD 6000 series were presented to the trade press in the form of the Radeon HD 6850 and 6870 on October 19, 2010, before the NDA officially fell three days later . Both cards use the Barts graphics processor, which is still based on the " Terascale 2 architecture " of the Radeon HD 5000 series and whose shader units are thus built in the 5D VLIW design. The cluster structure of the GPU (16 shader and four texture units) is also identical, which means that Barts has a total of 14 clusters of 224 shader and 56 texture units. Since AMD does not assign the Barts GPU to the high-end sector, double-precision support was dispensed with, which meant that transistors could be saved. Total Barts consists of 1.7 billion transistors on a The mm² face of 255th After Nvidia was able to demonstrate advantages in the area of ​​tessellation performance with its Fermi architecture , AMD revised the tesselator for the Barts GPU. The "Fixed Function Unit", the thread management and the data buffer have been improved, which means that with moderate use of tessellation, AMD has been able to double its performance compared to the Cypress GPU . However, the advantage does not apply to larger tessellation loads. In addition to the possibility of 3D playback, a third generation “ Unified Video Decoder ” is implemented, which supports hardware acceleration for more video formats.

The Radeon HD 6870 uses an uncut Barts GPU and thus achieves a performance that is between the Radeon HD 5850 and the Geforce GTX 470 . Two of the 14 shader clusters are deactivated on the Radeon HD 6850, which means that it roughly achieves the performance of the Geforce GTX 460 (1024) , which has been relatively successful in the performance sector to date. In some test reports, the Radeon HD 6850 was initially rated “too positive”, because all shader clusters were partially activated on the test samples and the corresponding cards thus achieved a performance that did not correspond to that of the cards available in stores. The naming and the AF filter were sometimes controversial in the trade press : Although the banding effect (abrupt texture transitions) from the Evergreen series has been largely eliminated, the AF filter also tended to flicker more, which, among other things. a. is due to a lowering of the default filter quality in the Catalyst driver, which was also done on the older models of the Radeon HD 5000 series with Cypress GPU. Ultimately, AMD took back the lowering of the default filter quality with the Catalyst drivers 11.1a and 11.2, which at least partially remedied the fact of the flickering AF filter. The naming caused confusion, since the Radeon HD 6850 and 6870 had a lower performance compared to the Radeon HD 5850 and 5870, although the name suggests the opposite.

On December 15, 2010, AMD introduced the Radeon HD 6950 and 6970. These are based on the Cayman graphics processor, for which a new shader design was developed. Since the R600 GPU, a 5D VLIW arrangement has been used as the shader architecture , in which five stream processors (ALU) are combined to form a shader. Four ALUs were responsible for the calculation of the RGB and the alpha value, the fifth acted as a T-Unit for special function functions such as sine and cosine calculations. The disadvantage with this arrangement was that if a value was not required for a clock cycle, the corresponding stream processor would then be idle. According to AMD, the load with this arrangement was around 70%. With the Cayman GPU, this arrangement was changed in favor of a 4D VLIW system. The T-Unit, for whose tasks three ALUs are now combined, was omitted. This is said to have improved utilization and reduced the number of transistors required. Another advantage of the new arrangement is that instead of the previous five, only four ALUs are required to calculate with double the accuracy . As a result, the DP value is now 25% of the SP value instead of 20% as before. After Nvidia was able to demonstrate clear advantages in tessellation performance with the Fermi architecture, AMD implemented the tessellator twice on the Cayman GPU. Ultimately, the Cayman graphics processor consists of around 2.64 billion transistors on a die area of ​​389 mm² and accommodates 1536 stream processors, 96 texture and 32 raster units in 24 shader clusters.

The presentation of the Radeon HD 6950 and 6970 was initially delayed, although AMD did not provide any information about the reasons. The Radeon HD 6970 had roughly the same performance as the Geforce GTX 570 . Since it had been suspected in advance due to the architecture changes that the Radeon HD 6970 would compete with the Geforce GTX 580 , it could not meet the sometimes very high expectations. The fact that the AF filter's tendency to flicker, which was already known from the Barts GPU, also occurred unchanged on the Radeon HD 6970 also contributed to this. The power consumption was also criticized, despite the new Powertune option, although it was not higher than that of the equally fast Geforce GTX 570. However, the Radeon cards of the previous generation often had a lower power consumption than their Nvidia competitors with the same performance. The Radeon HD 6970 showed advantages at very high resolutions, partly due to the large 2 GB video memory.

The 6950, which replaced the Radeon HD 5870 , was rated significantly more positively in the trade press than the Radeon HD 6970 . This was u. a. The fact that Nvidia initially had no competing card on offer, but also that it only performed around 10% less than the Radeon HD 6970, but consumed up to 30% less power. On the 6950, two shader clusters of the Cayman GPU were deactivated, leaving 1408 stream processors. However, it was possible to reactivate the deactivated clusters on cards from the first batch with a so-called bios flash , as these had not yet been processed with a laser cut. This activation was facilitated by the new dual bios, but led to the loss of warranty. After some board partners had already created their own variants of the Radeon HD 6950 with 1 GB of memory, AMD officially presented them on January 25, 2011. This was placed by AMD against the Geforce GTX 560 Ti launched on the same day .

On March 8, 2011, the last model based on the Cayman graphics processor followed, the Radeon HD 6990. Two Cayman GPUs were placed on this in full configuration. This dual GPU graphics card initially turned out to be the fastest card on the market when it was presented. However, the card also set new negative records for power consumption and noise generation. AMD installed two 8-pin power plugs in the 6990, increasing the TDP to 375 watts, thereby breaking the official PCI-SIG specifications. At the same time, AMD equipped the card with a second BIOS that increases the chip clock from 830 to 880 MHz in order to reach the clock level of the Radeon HD 6970. With the use of the alternative BIOS, called "AUSUM" by AMD (Antilles Unlocking Switch for Uber Mode), the warranty claim is lost and the TDP increases again to 450 watts (the power consumption can be up to 506 watts). Overall, the product presentation looked more like a feasibility study, because the higher the consumption, the higher the demands on the cooling system. However, these requirements are not met with the built-in cooler. The volume increases enormously compared to other products. With the alternative Bios "AUSUM", it is unreasonable to run the computer in the same room in which you are playing due to the permanently high level of noise pollution (sometimes with over 14 sons ).

On February 4, 2011, AMD presented the Radeon HD 6450, 6570 and 6670 models. The cards intended for the low-cost sector were initially only intended for the OEM market , with the Radeon HD 6450 based on the Caicos GPU. The Radeon HD 6570 and 6670, on the other hand, use the 716 mil. Transistors existing Turks GPU. On January 20, AMD presented the Radeon HD 6750 and 6770 cards, which are new editions of the Radeon HD 5750 and 5770 models , which were also only intended for the OEM market. On April 7th, AMD presented the Radeon HD 6450 unchanged for the retail market, although the actual sales start did not take place until April 19th. The Radeon HD 6570 and 6670 models were also released for the retail market on this day.

Graphics processors

Graphics chip	architecture	production			units						API support			Video processor	cutting stelle
		process	transis- interfere	The - area	ROPs	Unified shaders			Texture units		DirectX	OpenGL	OpenCL
						Stream processors	Shader units	Shader cluster	TAUs	TMUs
Caicos (RV910)	Terascale 2	40 nm	0370 million	067 mm²	04th	0160	032 × 5D- VLIW	02	08th	08th	11.0	4.4 (Linux Mesa 18+: 4.4+, 4.5 almost completely)	1.2 (Linux Mesa: 1.1 almost complete, 1.2 in progress)	UVD 3.0	PCIe 2.0
Turks (RV930)			0720 million	118 mm²	08th	0480	096 × 5D-VLIW	06th	24	24				UVD 3.0
Juniper (RV840)			1040 million	166 mm²	16	0800	160 × 5D-VLIW	10	40	40				UVD 2.2
Barts (RV940)			1700 million	255 mm²	32	1120	224 × 5D-VLIW	14th	56	56				UVD 3.0
Cayman (RV970)	Terascale 3		2640 million	389 mm²	32	1536	384 × 4D-VLIW	24	96	96				UVD 3.0

Naming

The Northern Island series uses the same naming system as has been used since the Radeon HD 3000 series . All graphics cards are labeled with "AMD Radeon HD" and an additional four-digit number that generally begins with a "6" (for the series). The second and third digits are used to subdivide into different models.

division

• HD 64xx: low-end
• HD 65xx, HD 66xx and 67xx: mainstream
• HD 68xx: performance
• HD 69xx: high-end

Due to the general drop in prices on the market and currency fluctuations, AMD's original classifications do not generally apply.

Model data

model	Official launch	Graphics processor (GPU)						Graphics memory
		Type	Active units				Chip clock (MHz)	Size ( MB )	Clock rate (MHz)	Type	Storage interface
			ROPs	Shader - cluster	Stream processors	Texture units
Radeon HD 6450	Feb 4, 2011	Caicos	4th	2	160	8th	625	1024	533-800	DDR3	64 bit
							750	512	1600 (800) - 1800 (900)	GDDR5
Radeon HD 6570	Feb 4, 2011	Turks	8th	6th	480	24	650	512-2048	900	DDR3	128 bit
								512-1024	1800 (900) - 2000 (1000)	GDDR5
Radeon HD 6670	Feb 4, 2011	Turks	8th	6th	480	24	800	512-1024	2000 (1000)	GDDR5	128 bit
Radeon HD 6750	Jan. 20, 2011	Juniper (Pro)	16	9	720	36	700	1024	2300 (1150)	GDDR5	128 bit
Radeon HD 6770	Jan. 20, 2011	Juniper (XT)	16	10	800	40	850	1024	2400 (1200)	GDDR5	128 bit
Radeon HD 6790	Apr 5, 2011	Barts (LE)	16	10	800	40	840	1024	2100 (1050)	GDDR5	256 bit
Radeon HD 6850	Oct 22, 2010	Barts (Pro)	32	12	960	48	775	1024	2000 (1000)	GDDR5	256 bit
Radeon HD 6870	Oct 22, 2010	Barts (XT)	32	14th	1120	56	900	1024	2100 (1050)	GDDR5	256 bit
Radeon HD 6930	Dec. 19, 2011	Cayman (LE)	32	20th	1280	80	750	2048	2500 (1250)	GDDR5	256 bit
Radeon HD 6950	Jan. 25, 2011	Cayman (Pro)	32	22nd	1408	88	800	1024	2500 (1250)	GDDR5	256 bit
	Dec 15, 2010							2048
Radeon HD 6970	Dec 15, 2010	Cayman (XT)	32	24	1536	96	880	2048	2750 (1375)	GDDR5	256 bit
Radeon HD 6990	March 8, 2011	2 × Cayman (Antilles)	2 × 32	2 × 24	2 × 1536	2 × 96	830	2 × 2048	2500 (1250)	GDDR5	2 × 256 bits

Hints

• The specified clock rates are those recommended or specified by AMD; the I / O clock is specified for the memory clock. However, the exact clock rate can deviate by a few megahertz due to different clock generators, and the final definition 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.
• The date indicated is the date of the public presentation, not the date of availability of the models.

Performance data

The following theoretical performance data result for the respective models:

model	Computing power via the stream processors in GFlops		Graphics processor / memory
			Polygon throughput in million triangles / s	Pixel fill rate in GPixel / s	Texel fill rate in GT / s	Bandwidth in GB / s
	SP (MAD)	DP (FMA)
Radeon HD 6450 (DDR3)	200	-	k. A.	2.5	5	8.5-12.8
Radeon HD 6450 (GDDR5)	240	-	k. A.	3	6th	25.6-28.8
Radeon HD 6570 (DDR3)	624	-	650	5.2	15.6	28.8
Radeon HD 6570 (GDDR5)	624	-	650	5.2	15.6	57.6-64
Radeon HD 6670	768	-	800	6.4	19.2	64
Radeon HD 6750	1008	-	700	11.2	25.2	73.6
Radeon HD 6770	1360	-	850	13.6	34.0	76.8
Radeon HD 6790	1344	-	840	13.4	33.6	134.4
Radeon HD 6850	1488	-	775	24.8	37.2	128
Radeon HD 6870	2016	-	900	28.8	50.4	134.4
Radeon HD 6930	1920	480	1500	24	60	153.6
Radeon HD 6950	2253	563	1600	25.6	70.4	160
Radeon HD 6970	2703	676	1760	28.2	84.5	176
Radeon HD 6990	2 × 2550	2 x 637.5	2 × 1660	2 × 26.6	2 x 79.7	2 × 160

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 series, as it is based on a different architecture that scales differently.

Power consumption data

The measured values ​​listed in the table refer to the pure power consumption of graphics cards that correspond to the AMD reference design. A special measuring device is required to measure these values; Depending on the measurement technology used and the given measurement conditions, including the program used to generate the 3D load, the values ​​can fluctuate between different devices. Therefore, measured value ranges are given here, each representing the lowest and highest measured values ​​from different sources.

model	Type	Consumption ( watt )			additional power plug
		TDP	Readings
			Idle	3D load
Radeon HD 6450 (DDR3)	Caicos	020th			no
Radeon HD 6450 (GDDR5)	Caicos	027	007th	032	no
Radeon HD 6570 (DDR3)	Turks	044			no
Radeon HD 6570 (GDDR5)	Turks	060			no
Radeon HD 6670	Turks	066	012	057	no
Radeon HD 6750	Juniper	086			1 × 6 pin
Radeon HD 6770	Juniper	108			1 × 6 pin
Radeon HD 6790	Barts	150	017 ... 18	122..128	2 × 6-pin
Radeon HD 6850	Barts	127	019..25	117..124	1 × 6 pin
Radeon HD 6870	Barts	151	018-20	145..156	2 × 6-pin
Radeon HD 6930	Cayman	225			2 × 6-pin
Radeon HD 6950	Cayman	200	020th	185	1 × 6-pin 1 × 8-pin
Radeon HD 6970	Cayman	250	022nd	267	1 × 6-pin 1 × 8-pin
Radeon HD 6990	2 × Cayman	375	037 ... 42	351..376	2 × 8 pin

Much more common than measuring the consumption of the graphics card is determining the power consumption of an entire system. For this purpose, a reference system is compiled in which the various graphics cards are installed; Then the measurement takes place directly at the socket with the help of an energy cost meter or a comparable device . However, the meaningfulness of the measured values ​​is limited: It is not clear what consumption comes from the graphics card and what can be ascribed to the rest of the PC system. With this measurement method, the difference in consumption between idle and 3D load operation does not only depend on the program with which the load was generated; the utilization and efficiency of the rest of the PC system including the power supply unit, mainboard and processor also influence the measured difference. Since the tested systems usually differ from your own PC system at home, the values ​​given there cannot be mapped to your own system. Only measurement data from otherwise identical systems are (to a limited extent) suitable for comparison with one another. Because of this dependency, total system measured values ​​are not listed in the table here. However, since they can give a better picture of the practical power consumption of a specific system with a specific graphics card, websites that made such measurements are listed under the web links .

Web links

Commons : AMD-Radeon-HD-6000-Series  - collection of pictures, videos and audio files

• Official website

Measurement of the power consumption of an entire system

• Measurements of the entire system at ComputerBase.de
• Measurements of the entire system at TweakPC

Individual evidence

1. ↑ Radeon HD 6870/6850 in the test: Tessellation. PC Games Hardware, October 22, 2010, accessed October 24, 2010 . 
2. ↑ ^{a b} Radeon HD 6870/6850 in the test: conclusion. PC Games Hardware, October 22, 2010, accessed October 24, 2010 . 
3. ↑ Badly coordinated Barts attack: Radeon HD-6850 pattern with too many shaders in circulation. Hardtecs4u, October 22, 2010, accessed October 24, 2010 . 
4. ↑ Radeon HD 6870/6850 in the test: image quality & anisotropic filter. PC Games Hardware, October 22, 2010, accessed October 24, 2010 . 
5. ↑ AMD's Radeon HD 5800/5900/6800 with setbacks in filter quality. 3DCenter.org, November 2, 2010, accessed November 7, 2010 . 
6. ↑ AMD Radeon HD 6850 and HD 6870 test: Texture flicker in the Radeon HD 6870/6850. TweakPC, October 30, 2010, accessed November 7, 2010 . 
7. ↑ Report: Radeon HD 6800 - Image quality: progress or regression? ComputerBase, October 30, 2010, accessed November 7, 2010 . 
8. ↑ Improved: AMD Radeon HD 6950 1 GB and Catalyst Hotfix 11.1a. HT4U.net, February 10, 2011, accessed February 22, 2011 . 
9. ↑ Report: AMD Catalyst 11.1 + 11.1a Hot-Fix. ComputerBase, February 5, 2011, accessed April 19, 2011 . 
10. ↑ ^{a b c} Test: AMD Radeon HD 6970 and HD 6950 - Technology 1 - The innovations. ComputerBase, December 15, 2010, accessed March 8, 2011 . 
11. ↑ ^{a b} test: AMD Radeon HD 6970 and HD 6950 - performance rating. ComputerBase, December 15, 2010, accessed March 8, 2011 . 
12. ↑ Test: AMD Radeon HD 6970 and HD 6950 - power consumption. ComputerBase, December 15, 2010, accessed March 8, 2011 . 
13. ↑ Radeon HD 6950 to Radeon HD 6970 activated (update). ComputerBase, December 27, 2010, accessed March 8, 2011 . 
14. ↑ Test: AMD Radeon HD 6990 - performance ratings. ComputerBase, March 8, 2011, accessed March 8, 2011 . 
15. ↑ Review: AMD Radeon HD 6990 - power consumption. ComputerBase, March 8, 2011, accessed March 8, 2011 . 
16. ↑ ^{a b c d} Radeon HD 6990 in the test (Antilles): Loudness and power consumption. PC Games Hardware, March 8, 2011, accessed March 8, 2011 . 
17. ↑ Review: AMD Radeon HD 6990 - Technical data. ComputerBase, March 8, 2011, accessed March 8, 2011 . 
18. ↑ Test: AMD Radeon HD 6990 - volume. ComputerBase, March 8, 2011, accessed March 8, 2011 . 
19. ↑ AMD introduces the Radeon HD 6450. ComputerBase, April 7, 2011, accessed April 7, 2011 . 
20. ↑ AMD Radeon HD 6570 and 6670 presented: Round 2 in the entry-level area. PC Games Hardware, April 19, 2011, accessed April 19, 2011 . 
21. ↑ ^{a b c d e f g h i j k l m n} AMD Radeon HD 6790 - Barts-Chip market cut? - Power consumption: idle & load. HardTecs4U, April 5, 2011, accessed April 5, 2011 . 
22. ↑ ^{a b} Radeon HD 6790 in the test: How good is AMD's answer to the Geforce GTX 550 Ti? - Loudness and power consumption. PC Games Hardware, April 5, 2011, accessed April 5, 2011 . 
23. ↑ ^{a b c d} Radeon HD 6870/6850 in the test: Loudness and power consumption. PC Games Hardware, October 22, 2010, accessed October 22, 2010 . 
24. ↑ ^{a b} Throttled truck: AMD Radeon HD 6990 in the test - power consumption: idle & load. HardTecs4U, March 8, 2011, accessed April 23, 2011 .