Nvidia GeForce 600 series

GeForce GTX 650 Ti in DirectCU-II design from Asus

The GeForce 600 series is a series of desktop - graphics chip company Nvidia and successor of the GeForce 500 series . With the GeForce 600 series, which is also known as the Kepler generation due to the new architecture, Nvidia introduced incomplete support for DirectX 11.1, because only six of the ten new functions are supported compared to DirectX 11.0.

description

history

GK104

Nvidia GF117 (N13M-GE-S-A2)

On March 22nd, 2012 Nvidia presented the GeForce GTX 680, the first graphics card of the GeForce 600 series, with which the new Kepler architecture was introduced. The GeForce GTX 680 is based on the GK104 graphics processor, which consists of 3.54 billion transistors , as well as 1536 stream processors and 128 texture units , which are organized in eight shader clusters. The GK104-GPU is manufactured in the 28 nm manufacturing process at TSMC and has a die area of 294 mm². The GK104 was originally planned as a graphics chip for the performance sector. a. from the reduced “double-precision” performance , but also from the fact that earlier beta drivers carried the GeForce GTX 680 under the name GeForce GTX 670 Ti. After Nvidia dropped the GK100 graphics processor in favor of the GK110, the GK104 also had to be used for the high-end sector, as the GK110 was only to be available for the Kepler refresh generation.

At its presentation, the GeForce GTX 680 had, depending on the resolution, around 10 percent higher performance than the Radeon HD 7970 from competitor AMD , and 30 to 35 percent compared to its predecessor, the GeForce GTX 580 . The GeForce GTX 680 was initially the fastest single-chip graphics card on the market. In the trade press, the card was largely rated positively, which, in addition to its performance, was primarily due to its energy efficiency, as it consumed less power than the slower Radeon HD 7970 and GeForce GTX 580. The reference cooler was also rated positively in relation to a card in the high-end area.

On April 29, 2012, Nvidia presented the dual-chip graphics card GeForce GTX 690, which was based on two GK104 GPUs in full configuration. The card showed around 40 percent higher performance than its predecessor, the GeForce GTX 590 . Since AMD did without the previously announced Radeon HD 7990 (only a few board partners published their own designs under this name), the GeForce GTX 690 assumed a monopoly position as the fastest graphics card on the market. Since Nvidia was able to significantly reduce the power consumption compared to earlier dual-chip graphics cards, due to the advantages of the Kepler architecture, but also the noise (due to the lower heat generation), the card was rated positively in the trade press. This was also due to the fact that Nvidia was able to reduce the phenomenon of "micro stuttering" that occurs in SLI mode (just like with AMD's Crossfire ).

On May 10, 2012, Nvidia presented the third graphics card based on the GK104 GPU, the GeForce GTX 670. With this, a shader cluster of the GK104 graphics processor was deactivated (with 1344 stream processors and 112 texture units still active), the chip clock was reduced and the reference - Revised circuit board layout. As a result, the GeForce GTX 670 lost about 10 percent performance compared to the GTX 680 and was about as fast as the Radeon HD 7970 (as long as no extreme resolutions above " Full HD " were used). Once again, Nvidia received positive reviews for the better energy efficiency compared to the AMD competitor, but the simplified reference cooler caused negative reviews, as it was louder compared to the GeForce GTX 680 despite the lower heat development. Most board partners reacted to the reference cooler with their own designs to counter the criticism.

On August 16, 2012, Nvidia introduced the GeForce GTX 660 Ti in the market. The card is technically identical to the GeForce GTX 670, only the memory interface has been reduced from 256 to 192 bits; The chip configuration, circuit board layout, clock rates and memory configuration are otherwise unchanged. The reduced memory interface and the associated asynchronous memory configuration meant that the GeForce GTX 660 Ti lost around 15 percent of its performance compared to the GTX 670 and thus had roughly the same performance as the Radeon HD 7950 . The GeForce GTX 660 Ti was the first graphics card with the GK104 GPU that Nvidia brought to market for less than 300 € and thus placed it in the important performance sector for which the GK104 was originally designed. For the OEM market , Nvidia launched another version of the GeForce GTX 660 Ti in which two shader clusters were deactivated and the clock rates were further reduced. This card was delivered under the name GeForce GTX 660 (without Ti), but should not be confused with the GeForce GTX 660 based on the GK106 graphics processor.

GK106

The GK106 graphics processor was introduced a good six months after the GK104 and GK107, which is probably due to a redesign from four to five shader clusters (this is what makes the GK106 the only Kepler GPU to have an asynchronous ratio of GPC to the SMX blocks in full expansion). Ultimately, the GK106 consists of 2.54 billion transistors on a die area of 214 mm² and has 960 stream processors and 80 texture units .

Nvidia presented the first graphics card based on the GK106 on September 13, 2012 with the GeForce GTX 660. The GeForce GTX 660 had a high market relevance for Nvidia, because after AMD had already presented the Radeon HD 7850 and 7870 in March 2012 , Nvidia could not offer any competitive offers in the important market segment around € 200. This situation presumably led to the redesign of the GK106, since in its original form with only four shader clusters it could not keep up with the AMD competition and Nvidia had to give up this market segment for over half a year. Ultimately, Nvidia placed the GeForce GTX 660 in terms of both performance and price between the AMD competitors Radeon HD 7850 and 7870. The reviews of the GeForce GTX 660 were very different in the trade press, which was partly due to the fact that Nvidia did not specify a reference design and thus the hardware testers received completely different test samples from different board partners, which differed both in terms of performance, but especially in terms of the cooler design (and thus the noise level). With the GeForce GTX 660, Nvidia again used the asynchronous memory configuration of 2 GB Vram and a 192 bit memory interface. In contrast to the GeForce GTX 660 Ti, in which performance drops were observed in some cases, there were no disadvantages on the GeForce GTX 660 due to this configuration in 3D applications.

Nvidia presented the second graphics card based on the GK106 on October 9, 2012 with the GeForce GTX 650 Ti. The naming caused confusion, as a GeForce GTX 650 had already been introduced before, but it was based on the significantly slower GK107 graphics processor. The GK106 on the GeForce GTX 650 Ti comes with a deactivated shader cluster, which means that four active clusters are still available. Furthermore, the memory interface has been reduced to 128 bits and the GPU boost function was missing. Nvidia placed the card against the Radeon HD 7770 , compared to which you could have around 10 percent higher performance with slightly lower power consumption. However, Nvidia set the starting price too high, which, in addition to the naming, caused criticism.

To close the relatively large performance gap between the GeForce GTX 660 and the GTX 650 Ti and to respond to the Radeon HD 7790 from AMD, Nvidia introduced the GeForce GTX 650 Ti Boost on March 26, 2013. Contrary to the name, the GeForce GTX 650 Ti Boost is not so much a drilled out GTX 650 Ti, but a GeForce GTX 660 in which an SMX block is deactivated (i.e. four out of five shader clusters are still active). The rest of the specifications are identical to those of the GeForce GTX 660. The card achieved roughly the same performance as the AMD competitor Radeon HD 7850.

GK107

The GK107 graphics processor is the smallest GPU of the Kepler generation. It only has two shader clusters and thus has 384 stream processors and 32 texture units . The GK107 consists of 1.3 billion transistors on a die area of 118 mm². No GK107 graphics card published to date supports the GPU boost function, although it is currently unclear whether this feature is generally not supported by the GK107 or whether Nvidia has not yet activated it. Nvidia initially only used the GK107 in the mobile sector before the GeForce GT 630 and GT 640 were presented on April 24, 2012. These cards, along with some new editions from the older Fermi generation, were only intended for the OEM market . It was not until June 5, 2012 that Nvidia also presented a GeForce GT 640 for the retail market, which was heavily criticized in the specialist press, which was due to the high price. The improved GeForce GTX 650 followed on September 13, 2012, in which Nvidia replaced the DDR3 with faster GDDR5 memory . Despite the performance improvements achieved, this card was also rated negatively because it was too expensive compared to the AMD competition.

GK208

The GK208 graphics processor is, in the broadest sense, a new revision of the GK107 GPU with only one grid partition. However, the chip has 512 kB cache per raster partition (also 512 kB in total) while GK107 still had 128 kB per raster partition (a total of 256 kB). Also there are actually only 8 TMUs per SMX and no more 16. In addition, some new compute features that were introduced with GK110 are supported (compute capability 3.5 instead of 3.0 as with the other GK10x chips).

In contrast to the other Kepler GPUs, Nvidia did not announce an official transistor value or die size (the latter can be estimated at around 90 mm² with the help of die screenshots). The GPU was initially only installed in the mobile sector before Nvidia presented the second revision of the GeForce GT 630 and 640 based on the GK208 on May 29, 2013. Nvidia is the only Kepler GPU to officially support DirectX 11.1 for the GK208, but with "Feature Level 11.0", which means that the exact support situation is unclear.

Rebranding

As usual with older graphics card generations, Nvidia repeated the practice of releasing graphics cards of the older generation under a new name with the GeForce 600 series. This process, which is primarily used for the OEM market , is known as "rebranding". Nvidia first relaunched the GeForce 510 and GT 520 as GeForce 605 and GT 620 on April 3, 2012 . On April 24th, the process was repeated with the GeForce GT 545 (DDR3) and GTX 555 as GeForce GT 640 and GT 645. On May 15, the GeForce GT 520 followed again as GeForce 610, as well as the GeForce GT 430, GT 440 (DDR3) and GT 440 (GDDR5) as GeForce GT 620, GT 630 (DDR3) and GT 630 (GDDR5).

architecture

The newly developed Kepler architecture, named after the German mathematician Johannes Kepler , which replaces the previous Fermi architecture, serves as the basis of the GeForce 600 series . Although Nvidia describes the Kepler architecture as a new development, it actually represents a further development of the previous Fermi generation. The primary change is the elimination of the "hot clock" for the shader units or stream processors. Since the unified architecture of the G80 graphics processor Nvidia used a so-called "hot clock", with which the shader units had a separate, higher clock than the rest of the GPU. This enabled Nvidia to achieve higher performance with fewer shader units, but this also meant that the GPUs had poorer energy efficiency than the graphics processors from competitor AMD. With the elimination of the "hot clock" Nvidia was able to fix this shortcoming and also block more shader units, as these now less space on the need (as an imaginary account indicates Nvidia that at a 10 percent lower power consumption 80 percent more die area for to use the shader units).

Independent of the revised stream processors, Nvidia has simplified the “scheduling” of the arithmetic commands and outsourced them to the software. Furthermore, the raster engine is now able to supply all ROPs with one pixel per cycle, which was not possible on the Fermi generation.

The basic structure of the Kepler architecture is comparable to Fermi: The GPUs are still made up of so-called “Graphics Processing Clusters”, or GPC for short (the GK104 consists of four GPCs as an example). Each GPC consists of two shader clusters (called SMX blocks by Nvidia), which in turn each accommodate 192 stream processors as well as load-and-store units and special function units. Furthermore, 16 instead of four to eight texture units are now built into each shader cluster , which can address and texture one pixel per cycle. The structure of the raster units remained unchanged: They are still divided into ROP clusters, with each cluster containing eight “raster operation processors” and a 64-bit memory controller.

The GPU boost function is a special feature of the Kepler architecture. This is a dynamic overclocking function that always speeds up the GPU if a certain “Power Target Limit” set by Nvidia is not reached. Other factors are power consumption, temperature and the currents applied. The graphics memory is not affected by this function.

Data overview

Graphics processors

Graphics chip	production			units						L2 cache	API support			Video pro- cessor	Bus interface stelle
	production process	transis- interfere	The - area	ROP particle functions	ROPs	Unified shaders		Texture units			DirectX	OpenGL	OpenCL
						Stream processors	Shader - cluster	TAUs	TMUs
GK104	28 nm	3.54 billion	294 mm²	4th	32	1536	8th	128	128	512 KB	11.0	4.5	1.2	VP5	PCIe 3.0
GK106		2.54 billion	214 mm²	3	24	0960	5	080	080	384 KB
GK107		1.30 billion	118 mm²	2	16	0384	2	032	032	256 KB
GK208		k. A.	087 mm²	1	08th	0384	2	016	016	512 KB					PCIe 2.0
GF108	40 nm	0.58 billion	114 mm²	1	04th	0096	2	016	016	k. A.		4.4	1.1	VP4
GF114		1.95 billion	358 mm²	4th	32	0384	8th	064	064	512 KB
GF116		1.17 billion	228 mm²	3	24	0192	4th	032	032	384 KB
GF119		0.29 billion	079 mm²	1	04th	0048	1	008th	008th	k. A.				VP5

Model data

model	Official launch	Graphics processor (GPU)								Graphics memory				Performance data
		Type	Active units				Chip clock (in MHz)			Size (in MB )	Clock rate (in MHz)	Type	Storage interface	Computing power ( GFlops )		Fill rate		Polygon - throughput (million triangles / s)	Memory bandwidth ( GB / s)
			ROPs	Shader - cluster	ALUs	Texture units	default	Shader	Boost					SP (MAD)	DP (FMA)	Pixels (GP / s)	Texel ( GT / s)
GeForce 605	0Apr 3, 2012	GF119	4th	1	48	8th	523	1046	-	512-1024	898	DDR3	64 bit	100.4	8.4	131	2.1	4.2	14.4
GeForce GT 610	May 15, 2012	GF119	4th	1	48	8th	810	1620	-	1024-2048	898	DDR3	64 bit	155.5	13	203	3.2	6.5	14.4
GeForce GT 620	0Apr 3, 2012	GF119	4th	1	48	8th	810	1620	-	512-1024	898	DDR3	64 bit	155.5	13	203	3.2	6.5	14.4
GeForce GT 620	May 15, 2012	GF108	4th	2	96	16	700	1400	-	1024	898	DDR3	64 bit	268.8	22.4	175	2.8	11.2	14.4
GeForce GT 630	Apr 24, 2012	GK107	16	1	192	16	875		-	1024-2048	891	DDR3	128 bit	336	14th	438	7th	14th	28.5
GeForce GT 630 DDR3	May 15, 2012	GF108	4th	2	96	16	810	1620	-	1024	898	DDR3	128 bit	311	25.9	203	3.2	13	28.8
GeForce GT 630 GDDR5	May 15, 2012	GF108	4th	2	96	16	810	1620	-	1024	800 (1600)	GDDR5	128 bit	311	25.9	203	3.2	13	51.2
GeForce GT 630	May 29, 2013	GK208	8th	2	384	16	902		-	2048	898	DDR3	64 bit	692.7	14.5	902	7.2	28.9	14.4
GeForce GT 635	19th Feb. 2013	GK208	8th	2	384	16	967		-	2048	1001	DDR3	64 bit	742.7	30.9	967	7.7	15.5	16
GeForce GT 640 (128-bit DDR3)	Apr 24, 2012	GK107	16	2	384	32	797		-	1024-2048	891	DDR3	128 bit	612.1	25.5	797	6.4	25.5	28.5
GeForce GT 640 (192-bit DDR3)	Apr 24, 2012	GF116	24	3	144	24	720	1440	-	1536-3072	891	DDR3	192 bits	414.7	34.6	540	8.6	17.3	42.8
GeForce GT 640 (128-bit GDDR5)	Apr 24, 2012	GK107	16	2	384	32	950		-	1024-2048	1250 (2500)	GDDR5	128 bit	729.6	30.4	950	7.6	30.4	80
GeForce GT 640	0Jun 5, 2012	GK107	16	2	384	32	900		-	1024	900	DDR3	128 bit	691.2	28.8	900	7.2	28.8	28.5
GeForce GT 640 GDDR5	May 29, 2013	GK208	8th	2	384	16	1046		-	1024	1250 (2500)	GDDR5	64 bit	803.3	16.8	1046	8.4	33.5	40
GeForce GT 645	Apr 24, 2012	GF114	24	6th	288	48	776	1552	-	1024	957 (1914)	GDDR5	192 bits	894	74.5	1164	18.6	37.2	91.9
GeForce GTX 645	Apr 22, 2013	GK106	16	3	576	48	823		-	1024	1000 (2000)	GDDR5	128 bit	948.1	39.5	823	13.2	39.5	64
GeForce GTX 650	13 Sep 2012	GK107	16	2	384	32	1053		-	1024	1250 (2500)	GDDR5	128 bit	808.7	33.7	1058	8.4	33.7	80
GeForce GTX 650 Ti	0Oct 9, 2012	GK106	16	4th	768	64	925		-	1024-2048	1350 (2700)	GDDR5	128 bit	1420.8	59.2	1829	14.8	59.2	86.4
GeForce GTX 650 Ti Boost	March 26, 2013	GK106	24	4th	768	64	980		1033	1024	1502 (3004)	GDDR5	192 bits	1505.3	62.7	1960	15.7-23.5	62.7	144.2
GeForce GTX 660	Aug 20, 2012	GK104	24	6th	1152	96	823		888	1536-3072	1450 (2900)	GDDR5	192 bits	1896.2	79	2469	19.8	79	134
GeForce GTX 660	13 Sep 2012	GK106	24	5	960	80	980		1033	2048	1502 (3004)	GDDR5	192 bits	1881.6	78.4	2450	23.5	78.4	144.2
GeForce GTX 660 Ti	16 Aug 2012	GK104	24	7th	1344	112	915		980	2048	1502 (3004)	GDDR5	192 bits	2459.5	102.5	3203	21.96	102.5	144.2
GeForce GTX 670	May 10, 2012	GK104	32	7th	1344	112	915		980	2048	1502 (3004)	GDDR5	256 bit	2459.5	102.5	3203	29.3	102.5	192.3
GeForce GTX 680	March 22, 2012	GK104	32	8th	1536	128	1006		1058	2048	1502 (3004)	GDDR5	256 bit	3090.4	128.8	4024	32.2	128.8	192.3
GeForce GTX 690	Apr 29, 2012	2 × GK104	2 × 32	2 × 8	2 × 1536	2 × 128	915		1019	2 × 2048	1502 (3004)	GDDR5	2 × 256 bits	2 x 2810.9	2 x 117.2	2 × 3660	2 x 29.3	2 x 117.2	2 x 192.3

Power consumption data

model	Type	Consumption ( watt )				additional power plug
		MGCP	Readings
			Idle	3D load	Maximum load
GeForce 605	GF119	025th	k. A.	k. A.	k. A.	no
GeForce GT 610	GF119	029	k. A.	k. A.	k. A.	no
GeForce GT 620 (OEM)	GF119	030th	k. A.	k. A.	k. A.	no
GeForce GT 620	GF108	049	k. A.	k. A.	k. A.	no
GeForce GT 630 (OEM)	GK107	050	k. A.	k. A.	k. A.	no
GeForce GT 630 DDR3	GF108	049	k. A.	k. A.	k. A.	no
GeForce GT 630 GDDR5	GF108	065	k. A.	k. A.	k. A.	no
GeForce GT 630	GK208	025th	k. A.	k. A.	k. A.	no
GeForce GT 640 (128-bit DDR3)	GK107	050	k. A.	k. A.	k. A.	no
GeForce GT 640 (192-bit DDR3)	GF116	075	k. A.	k. A.	k. A.	no
GeForce GT 640 (128-bit GDDR5)	GK107	075	k. A.	k. A.	k. A.	no
GeForce GT 640	GK107	065	8th	34	k. A.	no
GeForce GT 640 GDDR5	GK208	049	k. A.	k. A.	k. A.	no
GeForce GT 645	GF114	140	k. A.	k. A.	k. A.	1 × 6-pin
GeForce GTX 645	GK106	130	k. A.	k. A.	k. A.	1 × 6-pin
GeForce GTX 650	GK107	064	7-9	50-60	68	1 × 6-pin
GeForce GTX 650 Ti	GK106	110	6-9	63-73	84	1 × 6-pin
GeForce GTX 650 Ti Boost	GK106	134	k. A.	k. A.	k. A.	1 × 6-pin
GeForce GTX 660 (OEM)	GK104	130	k. A.	k. A.	k. A.	2 × 6-pin
GeForce GTX 660	GK106	140	7-10	112-118	138	1 × 6-pin
GeForce GTX 660 Ti	GK104	150	14-18	129-148	131-148	2 × 6-pin
GeForce GTX 670	GK104	170	14-15	144-147	162-184	2 × 6-pin
GeForce GTX 680	GK104	195	14-15	174-183	191-228	2 × 6-pin
GeForce GTX 690	2 × GK104	300	22-26	274-291	324-334	2 × 8-pin

Remarks

1. ↑ The date indicated is the date of the public presentation, not the date of availability of the models.
2. ↑ The specified performance values ​​for the computing power via the stream processors, the pixel and texel filling rate, as well as the memory bandwidth are theoretical maximum values ​​(with standard clock rate) that cannot be directly compared with the performance values ​​of other architectures. 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.
3. ↑ ^{a b} The specified clock rates are the reference data recommended or specified by Nvidia; 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.
4. ↑ ^{a b c d e f g h i j} OEM product. Card is not available in the retail market.
5. ↑ The MGCP value specified by Nvidia does not necessarily correspond to the maximum power consumption. This value is also not necessarily comparable with the TDP value of the competitor AMD.
6. ↑ The measured values ​​listed in the table relate to the pure power consumption of graphics cards that correspond to the Nvidia 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, typical and highest measured values ​​from different sources.
7. ↑ The value given under 3D load corresponds to the typical game usage of the card. However, this is different depending on the 3D application. As a rule, a modern 3D application is used to determine the value, which, however, limits the comparability over longer periods of time.
8. ↑ The maximum load is usually determined with demanding benchmark programs, the loads of which are significantly higher than those of "normal" 3D applications.

Web links

Commons : Nvidia GeForce 600 series  - collection of images, videos and audio files

• Product overview on the manufacturer's website

Individual evidence

1. ↑ Nvidia's Kepler GPUs are not fully compatible with DirectX 11.1. heise.de, November 21, 2012, accessed on November 9, 2013 . 
2. ↑ GeForce GT 640 and GT 630 for the first time with DirectX 11.1 (update). Computerbase, May 30, 2013, accessed November 9, 2013 . 
3. ↑ ^{a b c d e f} Test: Nvidia GeForce GTX 680 - architecture changes. ComputerBase, March 22, 2012, accessed October 11, 2012 . 
4. ↑ Review: Nvidia GeForce GTX 680 - Turbo mode. ComputerBase, March 22, 2012, accessed October 11, 2012 . 
5. ↑ http://www.geeks3d.com/20150519/nvidia-r352-86-whql-released-one-new-opengl-extension/ Support for OpenGL 4.5 and OpenCL 1.2 with driver 35x.xx and higher for Maxwell and Kepler
6. ↑ NVIDIA GeForce 605. Nvidia Corporation, accessed April 4, 2012 . 
7. ↑ NVIDIA GeForce GT 610. Nvidia Corporation, accessed November 9, 2013 . 
8. ↑ ASUS GT610-SL-2GD3-L. ASUS, accessed February 20, 2019 . 
9. ↑ NVIDIA GeForce GT 620. Nvidia Corporation, accessed April 4, 2012 . 
10. ↑ NVIDIA GeForce GT 620. Nvidia Corporation, accessed November 9, 2013 . 
11. ↑ NVIDIA GeForce GT 630. Nvidia Corporation, accessed April 24, 2012 . 
12. ↑ ^{a b c} NVIDIA GeForce GT 630. Nvidia Corporation, accessed November 9, 2013 . 
13. ↑ NVIDIA GeForce GT 635 (OEM). Nvidia Corporation, accessed November 9, 2013 . 
14. ↑ ^{a b c} NVIDIA GeForce GT 640 (OEM). Nvidia Corporation, accessed November 9, 2013 . 
15. ↑ ^{a b} NVIDIA GeForce GT 640. Nvidia Corporation, accessed June 5, 2012 . 
16. ↑ NVIDIA GeForce GT 645. Nvidia Corporation, accessed April 24, 2012 . 
17. ↑ NVIDIA GeForce GTX 645. Nvidia Corporation, accessed November 9, 2013 . 
18. ↑ NVIDIA GeForce GTX 650. Nvidia Corporation, accessed September 13, 2012 . 
19. ↑ NVIDIA GeForce GTX 650 Ti. Nvidia Corporation, accessed October 9, 2012 . 
20. ↑ NVIDIA GeForce GTX 650 Ti BOOST. Nvidia Corporation, accessed November 9, 2013 . 
21. ↑ NVIDIA GeForce GTX 660. Nvidia Corporation, accessed August 21, 2012 . 
22. ↑ NVIDIA GeForce GTX 660. Nvidia Corporation, accessed September 13, 2012 . 
23. ↑ NVIDIA GeForce GTX 660 Ti. Nvidia Corporation, accessed August 16, 2012 . 
24. ↑ NVIDIA GeForce GTX 670. Nvidia Corporation, accessed May 10, 2012 . 
25. ↑ NVIDIA GeForce GTX 680. Nvidia Corporation, accessed March 23, 2012 . 
26. ↑ NVIDIA GeForce GTX 690. Nvidia Corporation, accessed May 3, 2012 . 
27. ↑ ^{a b c} Small Kepler under 100 euros - EVGA GT 640 and GTX 650 SC in the test: power consumption. HardTecs4U, September 29, 2012, accessed October 10, 2012 . 
28. ↑ ^{a b c d e f g h i j k l m n o p q r} MSI GeForce GTX 650 Ti Power Edition 1 GB - Power Consumption. TechPowerUp, October 9, 2012, accessed October 10, 2012 . 
29. ↑ ^{a b c} GeForce GTX 650 Ti in the test: cooling, loudness and power consumption. PC Games Hardware, October 9, 2012, accessed October 10, 2012 . 
30. ↑ ^{a b c d e f} GeForce GTX 660 in the test: cooling, loudness and power consumption. PC Games Hardware, September 13, 2012, accessed October 10, 2012 . 
31. ↑ ^{a b c d} ASUS GTX 660 Ti DC2 Top - NVIDIA's GeForce GTX 660 Ti is here: power consumption - idle. HardTecs4U, August 16, 2012, accessed October 10, 2012 . 
32. ↑ ^{a b c} ASUS GTX 660 Ti DC2 Top - NVIDIA's GeForce GTX 660 Ti is here: Power consumption - games (Hawx). HardTecs4U, August 16, 2012, accessed October 10, 2012 . 
33. ↑ ^{a b c d} ASUS GTX 660 Ti DC2 Top - NVIDIA's GeForce GTX 660 Ti is here: power consumption - full load (Furmark). HardTecs4U, August 16, 2012, accessed October 10, 2012 .