Nvidia GeForce 400 series

GeForce GTX 480 from Point of View in reference design

The GeForce 400 series is a series of desktop - graphics chip company Nvidia . For the first time, all graphics processors in this series support the Shader model 5.0 (SM 5.0) according to DirectX 11, as well as OpenCL , CUDA and thus also PhysX .

description

history

GF100 graphics processor in A3 stepping

With the GeForce 400 series, Nvidia introduced support for the Shader model 5.0 based on DirectX 11. Originally, it was supposed to be presented in the 4th quarter of 2009, to coincide with the start of Windows 7 (with which DirectX 11 was released). In contrast to its competitor AMD , Nvidia decided to introduce a completely new architecture, which, however, required more development time. As a result, AMD was able to bring the full Radeon HD 5000 series line-up to market before development of the first GeForce 400 series graphics processor was completed. This was the GF100 manufactured in the 40 nm manufacturing process, which was the first to be based on the Fermi architecture. The GPU, which consists of around three billion transistors , was presented by Nvidia on January 18, 2010, but without presenting the corresponding graphics cards. As a result of the architecture changes , the number of stream processors on the GF100 per cluster has increased from 24 to 32 compared to its predecessor GT200 . Since the GF100 has a total of 16 shader clusters, 512 stream processors are available to it. Due to the new ratio of shaders to TMUs of 8: 1, the number of texture units is reduced from 80 to 64. The GF100 also has 48 raster operation processors, which are divided into six partitions. Each ROP partition is attached to a 64-bit memory controller for GDDR5 memory , resulting in a 384-bit memory interface. This enables the memory to be expanded to 1.5 GB and 3 GB. In theory, it can also be expanded to 6 GB.

At the beginning of February 2010, Nvidia announced that the names for the first graphics cards based on the GF100 will be GeForce GTX 470 and GTX 480. Previously it was generally expected that the cards would be assigned to the GeForce 300 series . The official launch took place on March 27, 2010. Although the GeForce GTX 480 turned out to be the fastest single GPU card on the market when it was launched, it was criticized in the specialist press for its high power consumption, with the GeForce GTX 480 setting a new negative record, which is why it is also under the The nickname "Thermi" or also known as "Grillforce", based on the name of the Fermi architecture used. This also resulted in critical values ​​in the area of ​​temperature and noise development. It was unexpected for observers that the GF100 graphics processor on the GeForce GTX 480 is operated with a deactivated shader cluster, which is probably due to problems with the 40 nm manufacturing process at TSMC . On the other hand, two shader clusters are deactivated on the GeForce GTX 470, with which it achieved the performance of the AMD competitor Radeon HD 5870 presented 6 months earlier . Since it had better values ​​than the GeForce GTX 480 in terms of power consumption and noise development, the GTX 470 also received better reviews in the trade press. Both models were officially launched on April 12, 2010, before the GeForce GTX 465 followed on May 31, 2010. This continued to use a GF100 GPU, but with five deactivated shader clusters and a 256 bit memory interface. This places the GeForce GTX 465 between the AMD competitors Radeon HD 5830 and 5850 in terms of performance .

On July 12, 2010, Nvidia introduced the GeForce GTX 460, the first card based on the GF104 graphics processor. Compared to the GF100, Nvidia halved the number of shader clusters in the GF104 and reduced the ROPs to four partitions, which means that a maximum of one 256-bit memory interface can be installed. At the same time, there are now 48 instead of 32 stream processors per cluster, whereby the number of TMUs and SFUs has also doubled. Since the GF104 is not intended for products of the Quadro and Teslaser series , Nvidia reduced the possibilities for use in the area of GPU computing . The computing power has been massively reduced with twice the accuracy , which is, however, irrelevant for 3D applications. This saved around a billion transistors in the GF100. This contributed to the fact that the power consumption as well as the heat and noise development on the GeForce GTX 460 were again significantly lower than was the case with the criticized cards with the GF100. Nvidia presented the GTX 460 in two memory configurations: 768 and 1024 MB Vram. In terms of performance, the version with 768 MB Vram placed itself between the Radeon HD 5830 and the GeForce GTX 465. The expansion level with 1024 MB, on the other hand, was able to put itself in front of the GeForce GTX 465, although the naming suggests something different. Despite the official price recommendations of US $ 199 and US $ 229 for the release, Nvidia assigned the cards with the abbreviation “GTX” to the high-end sector. For the OEM market, Nvidia brought a version of the GeForce GTX 460 with reduced clock speeds, as well as a "Second Edition" or "Special Edition" in which another shader cluster was deactivated and a second version of the 1024 MB Vram variant (as GTX 460 v2), in which the clock rates increased, but the memory interface was reduced to 192 bits. This means that the memory controllers are equipped asynchronously, as is the case with the GeForce GTX 550 Ti .

On September 13, 2010, Nvidia introduced the GeForce GTS 450. This uses the GF106 graphics processor, which with 192 shader and 32 texture units represents a halved version of the GF104 GPU in the broadest sense. Although the GF106 has three grid partitions, with which a 192 bit memory interface would be possible, Nvidia only uses 128 bits in the GTS 450. This means that the memory in the reference design is 1024 MB, with 512 and 2048 MB also possible. In terms of 3D performance, the GeForce GTS 450, which has been praised in the trade press like the GeForce GTX 460 for its low noise level, is roughly the same as the Radeon HD 5750 . In direct comparison with this one, it achieves better values ​​in the area of ​​power consumption in idle, whereas it consumes less power under load. For the OEM market, Nvidia brought out an adapted version of the GeForce GTS 450 in which a shader cluster was deactivated, but the memory was increased to 1536 MB. Another modification of the GTS 450 for the OEM sector is the GeForce GT 440, which also has to do without a shader cluster and uses DDR3 memory instead of GDDR5 . In February 2011, Nvidia released the GT 440 for the retail market, but it was based on the GF108 graphics processor of the GeForce GT 430.

On October 11, 2010, Nvidia presented the GeForce GT 430. It was based on the GF108 graphics processor, which is a halved version of the GF106 GPU, putting the GT 430 between the AMD competitors Radeon HD 5550 and 5570 in terms of 3D performance placed. The card , which was produced in the low-profile format , was aimed primarily at "casual gamers" or for use in multimedia and HTPCs . Compared to the AMD competitors, it has a higher power consumption under load, but has advantages when playing Blu-ray media . On September 3, 2010, Nvidia listed the GeForce GT 420 for the OEM market on its website at the same time as the launch of a number of products from the GeForce 400M series . A few days later, it was removed from the website, which initially left it unclear whether the card was really "launched". Since the presentation of the GeForce GT 430, the GT 420 has also been listed again.

Fermi architecture

With the GeForce 400 series, Nvidia uses the newly developed "Fermi architecture" for the first time, which is also used on the Quadro and Tesla cards . Fermi is the successor to the unified shader architecture of the G80 graphics processor . The primary improvements relate to the support of DirectX 11 as well as the expanded application possibilities in the area of GPU computing .

The graphics processors based on the “Fermi architecture” primarily consist of the “Graphics Processing Clusters” (GPC). In addition to the “Raster Engine”, these “Graphics Processing Clusters” also accommodate four Shader Clusters or “Streaming Multiprocessors”. Each shader cluster in turn has 32 to 48 stream processors , four to eight texture units and a "PolyMorph engine". There are also 16 “load / store” units each, which calculate the source and destination addresses of 16 threads in one cycle and can write the results to the cache or VRAM. There are also four to eight "Special Function Units" (SFU) for calculating the sine and cosine . Each SFU can execute one instruction per thread per cycle , with eight cycles being required for a warp.

The "CUDA Cores" (an allusion to the CUDA API from Nvidia) are simple-scalar stream processors, which are still made up of a full-fledged "Arithmetic Logic Unit" (ALU) and a "Floating Point Unit" ( FPU). To improve GPU computing capabilities, the graphics processors of the "Fermi architecture" are the first to have complete support for C ++ and, just like the Radeon HD 5000 series from AMD , are compatible with the IEEE 754-2008 Standard fully compatible. The latter became necessary in order to be able to use FMA (Fused Multiply-Add), which is more precise than MAD , to improve double-precision capabilities (calculation with double precision ). Each stream processor can calculate a fused multiply ADD (FMA) per clock, regardless of whether it is a single-precision or a double-precision operation . In contrast to the previous generation, multiplication operations (MUL) are no longer possible on the "Fermi architecture".

So far, the texture units of the G80 and GT200 were combined in so-called “Texture Processing Clusters”. With the “Fermi architecture” this cluster is completely eliminated. Instead, there are four to eight texture units per shader cluster. As a result, the ratio of shaders to TMUs deteriorates to 8: 1 or 6: 1 (previously 2: 1 or 3: 1), but there is now a dedicated 12 KB L1 texture cache per shader cluster.

The Raster Operation Processors (ROP) have been partially reorganized in the Fermi architecture. As before, these are combined in partitions which are still attached to the memory controller , with up to eight raster output stages per partition. An ROP can output a 32-bit integer pixel after one cycle, a 16-bit floating point pixel after two cycles or a 32-bit FP pixel after four cycles. The maximum number of pixels that can be processed is limited by the fact that each shader cluster can only pass 2 (GF100) or 4 pixels (GF104, GF106 and GF108) to the ROPs per cycle. With the Fermi models that have appeared so far, the full number of ROPs can therefore only be used for the predominant processing of 16- and 32-bit floating point pixels, which limits the maximum pixel fill rate. When using formats higher than 32-bit pixels, however, not all ROPs can be fully utilized due to different data path allocation. However, this restriction does not apply to the Z fill rate.

In order to improve the GPU computing capabilities, the “Fermi architecture” has an L1 and L2 cache in addition to the shared memory . Each shader cluster has a 76 KB cache, with 12 KB L1 texture cache being specialized for the texture units. The remaining 64 KB are freely configurable, so that either the L1 cache 48 KB and the shared memory cache 16 KB, or vice versa. In addition, the “Fermi architecture” has a global L2 cache designed as “Unified”, which is 128 KB per memory controller and thus contains a total of 768 KB in the case of the GF100 (GT200: 256 KB). The unified design makes it possible to do without the L2 texture cache, an ROP cache and the on-chip FIFOs of earlier architectures. The L2 cache is responsible for receiving all load, store and texture requests, and all units can now access them at the same time.

Nvidia has reorganized the rendering pipelines for the "Fermi architecture". The GPU initially receives the commands from the CPU via the so-called host interface. The "GigaThread Engine" then copies the data from the system memory to its own video memory and divides it into thread blocks. These are then forwarded to the shader clusters via the "Graphics Processing Clusters" or their "Raster Engine", which Nvidia now also calls "Streaming Multiprocessors". Each block is now divided into 32 threads or warps, with each shader cluster being able to process 48 warps before they are forwarded to the stream processors .

Naming

The GeForce 400 series uses the naming scheme that was first introduced with the GeForce 200 desktop series. All graphics chips are identified with a letter abbreviation to classify the performance sector and a three-digit number that generally begins with a "4" (for GeForce 400). The last two digits serve for further differentiation within the respective service sector.

Letter abbreviation:

• GT or no prefix - low budget
• GTS - mainstream
• GTX - high-end and performance

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

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
GF100	40 nm	3.04 billion	526 mm²	6th	48	512	16	64	64	768 KB	11.0	4.4	1.1	VP4	PCIe 2.0
GF104		1.95 billion	332 mm²	4th	32	384	08th	64	64	512 KB
GF106		1.17 billion	238 mm²	3	24	192	04th	32	32	384 KB
GF108		0.58 billion	114 mm²	1	04th	096	02	16	16	k. A.
GF114		1.95 billion	332 mm²	4th	32	384	08th	64	64	512 KB
GF116		1.17 billion	238 mm²	3	24	192	04th	32	32	384 KB
GT216		0.49 billion	100 mm²	2	08th	048	02	16	16	k. A.	10.1	3.3
GT218		0.26 billion	057 mm²	1	04th	016	01	08th	08th	k. A.

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 (in GFlops )		Polygon throughput (in million triangles / s)	Pixel fill rate (in GPixel / s)	Texel fill rate (in GTexel / s)	Memory bandwidth (in GB / s)
			ROPs	Shader - cluster	ALUs	Texture units	GPU	Shader					SP (MAD)	DP (FMA)
GeForce 405	3rd Sep 2010	GT218	4th	1	16	8th	589	1402	512	790	DDR3	64 bit	44.9	-	k. A.	2.4	4.7	12.6
		GT216	8th	2	48	16	475	1100	512	800	DDR3	64 bit	44.9	-	k. A.	3.8	7.6	12.8
GeForce GT 420	3rd Sep 2010	GF108	4th	1	48	4th	700	1400	2048	900	DDR3	128 bit	134.4	11.2	350	1.4	2.8	28.8
GeForce GT 430	Oct 11, 2010	GF108	4th	2	96	16	700	1400	1024	800	DDR3	64 bit	268.8	22.4	350	2.8	11.2	12.8
GeForce GT 430	Oct 11, 2010	GF108	4th	2	96	16	700	1400	2048	800	DDR3	128 bit	268.8	22.4	350	2.8	11.2	25.6
GeForce GT 440	Feb. 1, 2011	GF108	4th	2	96	16	810	1620	1024	900	DDR3	128 bit	311	25.9	404	3.2	13	28.8
									512-1024	1600 (800)	GDDR5							51.2
GeForce GT 440	Oct 11, 2010	GF106	24	3	144	24	594	1189	1536	900	DDR3	192 bits	342.4	28.5	594	7.1	14.3	43.2
									3072	800								38.4
GeForce GTS 450	13 Sep 2010	GF106	16	4th	192	32	783	1566	1024	1804 (902)	GDDR5	128 bit	601.3	50.1	783	6.3	25.1	57.7
GeForce GTS 450 Rev. 2	March 15, 2011	GF116	16	4th	192	32	783	1566	1024	1804 (902)	GDDR5	128 bit	601.3	50.1	783	6.3	25.1	57.7
GeForce GTS 450 Rev. 3	Jul 11, 2012	GF116	16	3	144	24	783	1566	1024	1400 (700)	DDR3	128 bit	451	37.6	783	4.7	18.8	22.4
GeForce GTS 450	Oct 11, 2010	GF106	24	3	144	24	790	1580	1536	2000 (1000)	GDDR5	192 bits	455	37.9	790	4.7	19.0	96
GeForce GTX 460 768 MB	July 12, 2010	GF104	24	7th	336	56	675	1350	768	1800 (900)	GDDR5	192 bits	907.2	75.6	1350	9.5	37.8	86.4
GeForce GTX 460 1024 MB	July 12, 2010	GF104	32	7th	336	56	675	1350	1024	1800 (900)	GDDR5	256 bit	907.2	75.6	1350	9.5	37.8	115.2
GeForce GTX 460 SE	Nov 15, 2010	GF104	32	6th	288	48	650	1300	1024	1700 (850)	GDDR5	256 bit	748.8	62.4	1300	7.8	31.2	108.8
GeForce GTX 460 Rev. 2	Sep 24 2011	GF114	24	7th	336	56	778	1556	1024	2004 (1002)	GDDR5	192 bits	1045.6	87.1	1556	10.9	43.6	96.2
GeForce GTX 460	Aug 2, 2010	GF104	32	7th	336	56	650	1300	1024	1700 (850)	GDDR5	256 bit	873.6	72.8	1300	9.1	36.4	108.8
GeForce GTX 465	May 31, 2010	GF100	32	11	352	44	607	1215	1024	1604 (802)	GDDR5	256 bit	855.4	106.9	1821	13.4	26.7	102.7
GeForce GTX 470	March 27, 2010	GF100	40	14th	448	56	607	1215	1280	1676 (838)	GDDR5	320 bits	1088.6	136.1	2428	17.0	33.9	134.1
GeForce GTX 480	March 27, 2010	GF100	48	15th	480	60	700	1401	1536	1848 (924)	GDDR5	384 bits	1345	168.1	2800	21.0	42.0	177.4

Power consumption data

model	Type	Consumption ( watt )				additional power plug
		MGCP	Readings
			Idle	3D load	Maximum load
GeForce 405 (OEM)	GT218	25th	k. A.	k. A.	k. A.	no
GeForce GT 420 (OEM)	GF108	k. A.	k. A.	k. A.	k. A.	no
GeForce GT 430	GF108	49	7th	k. A.	48	no
GeForce GT 430 (OEM)	GF108	60	k. A.	k. A.	k. A.	no
GeForce GT 440	GF106	65	k. A.	k. A.	k. A.	no
GeForce GT 440 (OEM)	GF106	56	k. A.	k. A.	k. A.	no
GeForce GTS 450	GF106	106	12-15	k. A.	102-105	1 × 6 pin
GeForce GTS 450 Rev. 2	GF116	106	k. A.	k. A.	k. A.	1 × 6 pin
GeForce GTS 450 Rev. 3	GF116	106	k. A.	k. A.	k. A.	1 × 6 pin
GeForce GTS 450 (OEM)	GF106	106	k. A.	k. A.	k. A.	1 × 6 pin
GeForce GTX 460 768 MB	GF104	150	14th	k. A.	137	2 × 6-pin
GeForce GTX 460 1024 MB	GF104	160	16	k. A.	158	2 × 6-pin
GeForce GTX 460 SE	GF104	k. A.	k. A.	k. A.	k. A.	2 × 6-pin
GeForce GTX 460 Rev. 2	GF114	160	k. A.	k. A.	k. A.	2 × 6-pin
GeForce GTX 460 (OEM)	GF104	160	k. A.	k. A.	k. A.	2 × 6-pin
GeForce GTX 465	GF100	200	26-29	k. A.	181-204	2 × 6-pin
GeForce GTX 470	GF100	215	30-31	k. A.	231-239	2 × 6-pin
GeForce GTX 480	GF100	250	45-49	k. A.	304-318	1 × 6-pin 1 × 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} 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 400 series  - collection of images, videos and audio files

• Product overview on the manufacturer's website
• GF100 graphics processor on the official Nvidia homepage

Individual evidence

1. ↑ Nvidia GF100 is called GTX 470 and GTX 480. Hardware information, February 2, 2010, accessed on June 21, 2010 . 
2. ↑ Nvidia introduces GeForce GTX 480 and 470. ComputerBase, March 27, 2010, accessed June 21, 2010 . 
3. ↑ Test: Nvidia GeForce GTX 480 - performance rating. ComputerBase, March 27, 2010, accessed June 21, 2010 . 
4. ↑ http://www.pcgameshardware.de/Grafikkarten-Grafikkarte-97980/News/Von-G80-und-Co-838364/
5. ↑ Archived copy ( Memento of the original from March 4, 2016 in the Internet Archive ) Info: The archive link was inserted automatically and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice. 
6. ↑ Test: Nvidia GeForce GTX 480 - power consumption. ComputerBase, March 27, 2010, accessed June 21, 2010 . 
7. ↑ Test: Nvidia GeForce GTX 480 - volume. ComputerBase, March 27, 2010, accessed June 21, 2010 . 
8. ↑ Test: Nvidia GeForce GTX 480 - temperature. ComputerBase, March 27, 2010, accessed June 21, 2010 . 
9. ↑ GeForce GTX 470 and GTX 480: Review of Nvidia's GF100 generation - conclusion. PC Games Hardware, April 14, 2010, accessed June 21, 2010 . 
10. ↑ We asked Nvidia about the delivery situation, volume and power consumption of the GTX 470 and GTX 480. PC Games Hardware, April 21, 2010, accessed on June 21, 2010 . 
11. ↑ Test: Nvidia GeForce GTX 465 - performance rating. ComputerBase, May 31, 2010, accessed June 21, 2010 . 
12. ↑ ^{a b} GeForce GTX 460 in the test: Introduction and specs. PC Games Hardware, July 12, 2010, accessed July 12, 2010 . 
13. ↑ Review: Nvidia GeForce GTX 460 - power consumption. ComputerBase, July 12, 2010, accessed July 12, 2010 . 
14. ↑ ^{a b} test: Nvidia GeForce GTX 460 - performance rating. ComputerBase, July 12, 2010, accessed July 12, 2010 . 
15. ↑ NVIDIA GeForce GTX 460 - Introducing the Hunter (page 35). HardTecs4U, July 12, 2010, accessed July 12, 2010 . 
16. ↑ ^{a b} Test: Nvidia GeForce GTS 450 (SLI) - Technical data. ComputerBase, September 13, 2010, accessed September 13, 2010 . 
17. ↑ Test: Nvidia GeForce GTS 450 (SLI) - volume. ComputerBase, September 13, 2010, accessed September 13, 2010 . 
18. ↑ Test: Nvidia GeForce GTS 450 (SLI) - performance rating. ComputerBase, September 13, 2010, accessed September 13, 2010 . 
19. ↑ ^{a b c d e f g h i j k l} GeForce GTS 450 in the test - NVIDIA presents the quiet "Sniper" (page 18). HardTecs4U, September 13, 2010, accessed on September 13, 2010 . 
20. ↑ Nvidia introduces the GeForce GT 440 for end customers. ComputerBase, February 1, 2011, accessed February 1, 2011 . 
21. ↑ Test: Nvidia GeForce GT 430 - Ratings. ComputerBase, October 11, 2010, accessed October 14, 2010 . 
22. ↑ Review: Nvidia GeForce GT 430 - conclusion. ComputerBase, October 11, 2010, accessed October 14, 2010 . 
23. ↑ Report: Nvidia GF100: Technical Details - Streaming Multiprocessor (SM). ComputerBase, January 18, 2010, accessed June 21, 2010 . 
24. ↑ Fermi GF100 in the technical TÜV: Comments on architectural details, image quality and benchmarks. PC Games Hardware, January 18, 2010, accessed June 21, 2010 . 
25. ↑ Report: Nvidia GF100: Technical details - ROPs and memory interface. ComputerBase, January 18, 2010, accessed June 21, 2010 . 
26. ↑ Test: Nvidia GeForce GTX 470. ComputerBase, April 12, 2010, accessed on June 21, 2010 . 
27. ↑ GeForce GTX 470/480: Fill rate puzzle solved. PC Games Hardware, April 7, 2010, accessed June 21, 2010 . 
28. ↑ NVIDIA GF100 becomes more concrete - page 6: GF100 and the memory. Hardwareluxx, January 18, 2010, accessed June 21, 2010 . 
29. ↑ Report: Nvidia GF100: Technical Details - Caches in the GF100. ComputerBase, January 18, 2010, accessed June 21, 2010 . 
30. ↑ NVIDIA GeForce 405. Nvidia Corporation, accessed May 17, 2011 . 
31. ↑ NVIDIA GeForce GT 420 OEM. Nvidia Corporation, accessed September 3, 2010 . 
32. ↑ NVIDIA GeForce GT 430. Nvidia Corporation, accessed October 11, 2010 . 
33. ↑ NVIDIA GeForce GT 430 OEM. Nvidia Corporation, accessed October 14, 2010 . 
34. ↑ NVIDIA GeForce GT 440. Nvidia Corporation, accessed February 1, 2011 . 
35. ↑ NVIDIA GeForce GT 440 OEM. Nvidia Corporation, accessed October 22, 2010 . 
36. ↑ ^{a b c} NVIDIA GeForce GTS 450. Nvidia Corporation, accessed September 13, 2010 . 
37. ↑ NVIDIA GeForce GTS 450 OEM. Nvidia Corporation. Retrieved October 12, 2010 . 
38. ↑ ^{a b c d} NVIDIA GeForce GTX 460. Nvidia Corporation, accessed July 12, 2010 . 
39. ↑ NVIDIA GeForce GTX 460 OEM. Nvidia Corporation, accessed August 12, 2010 . 
40. ↑ NVIDIA GeForce GTX 465. Nvidia Corporation, accessed July 12, 2010 . 
41. ↑ NVIDIA GeForce GTX 470. Nvidia Corporation, accessed June 21, 2010 . 
42. ↑ NVIDIA GeForce GTX 480. Nvidia Corporation, accessed June 21, 2010 . 
43. ↑ ^{a b} GeForce GT 430 in the test: power consumption. PC Games Hardware, October 11, 2010, accessed October 11, 2010 . 
44. ↑ ^{a b} Nvidia GeForce GTS 450 in the test: What good is DirectX 11 for 130 euros? - Impressions, loudness and power consumption. PC Games Hardware, September 13, 2010, accessed September 13, 2010 . 
45. ↑ ^{a b c d} Review GeForce GTX 465: Nvidia's cheapest DirectX 11 graphics card - impressions, loudness and power consumption. PC Games Hardware, May 31, 2010, accessed June 21, 2010 . 
46. ↑ ^{a b c} GeForce GTX 470 and GTX 480: Test of the GF100 generation with SLI - impressions, loudness and power consumption. PC Games Hardware, April 14, 2010, accessed June 21, 2010 .