GeForce 600 Series
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 called due to the new architecture as Kepler generation, Nvidia introduced a partial support for DirectX 11.1, because only six of the ten newly added features compared to DirectX 11.0 are supported.
- 2.1 GPUs
- 2.2 Model data
- 2.3 Performance data recording
Description
History
GK104
On March 22, 2012, Nvidia unveiled the GeForce GTX 680, the first graphics card in the GeForce 600 series, with which the new Kepler architecture was introduced. The GeForce GTX 680 is based on the GK104 GPU, which consists of 3.54 billion transistors and 1536 stream processors and 128 texture units, which are organized into eight shader clusters. The GK104 GPU is manufactured in the 28nm manufacturing process at TSMC and comes on a die area of 294 mm ². Originally, the GK104 had been planned as a graphics chip for the performance sector, which, inter alia, to at the reduced " double precision " performance suggests, but also because the earlier beta drivers the Geforce GTX 680 led under the name GeForce GTX 670 Ti. After Nvidia dropped the GK100 GPU in favor of the GK110, GK104 which had to be used for the high-end sector, as the GK110 is only available for the Kepler refresh generation available.
The Geforce GTX 680 pointed at her presentation one, depending on the resolution, a 10 percent increase in performance over the Radeon HD 7970 from its competitor AMD, as well as 30 to 35 percent compared to its predecessor, the GeForce GTX 580 This was the Geforce GTX 680 first the fastest single-chip graphics card on the market. In the trade press card was rated predominantly positive, which was due not only to the performance of energy efficiency before all things, as they face the slower Radeon HD 7970 and GeForce GTX 580 had a lower power consumption. Also, the reference cooler was positively evaluated in relation to a map in the high-end range.
On April 29, 2012, Nvidia introduced the dual-chip graphics card Geforce GTX 690, which was based on two GK104 GPUs in full expansion. The card had a higher by about 40 percent performance against its predecessor, the GeForce GTX 590, on. Since AMD on the previously announced Radeon HD 7990 waived (only a few board partners own designs published under this name ), took the Geforce GTX 690 as the fastest graphics card on the market in a monopoly position. Since Nvidia was able to reduce the power consumption over previous dual-chip graphics card significantly due to the advantages of the Kepler architecture, but also the noise ( due to the lower heat generation), the card in the trade press has been positive. This was due to the fact that Nvidia the phenomenon of " micro stuttering " that arises in SLI mode could ( just as with AMD's Crossfire ), reduce.
10 May 2012 Nvidia presented the third graphics card based on the GK104 GPU, the GeForce GTX 670 This was a shader cluster of GK104 GPU disabled ( which still 1344 stream processors and 112 texture units were active ), the core clock speed is reduced and the reference - board layout revised. This meant that the GeForce GTX 670 compared to the GTX 680 lost about 10 percent in performance and was about as fast as the Radeon HD 7970 (as long as no extreme resolutions have been used for "Full HD" ). Again Nvidia received positive reviews for better energy efficiency compared to the AMD competitors, but made the simplified reference cooler for negative reviews, because it was louder despite the lower heat generation compared to the GTX 680. Most board partners responded with intrinsic designs on the reference cooler, to counter the criticism.
On August 16, 2012, Nvidia Geforce GTX 660 Ti introduced the one in the market. The map is technically with the Geforce GTX 670 is identical, only the memory interface was reduced from 256 to 192 bits; Chip development, PCB layout, clock speeds and memory configuration are otherwise unchanged. The reduced memory interface and the asynchronous memory configuration related meant that the GeForce GTX 660 Ti over the GTX 670 about 15 percent lost power and thus about the performance of the Radeon HD 7950 had. The Geforce GTX 660 Ti was the first graphics card with the GK104 GPU, the Nvidia brought for less than 300 € on the market and thus placed in the key performance sector, for which the GK104 was originally designed. Nvidia designed for the OEM market to another variant of the Geforce GTX 660 Ti, in which two shader clusters were disabled and the clock rates have been further reduced. This card has been delivered under the name GeForce GTX 660 (without Ti), but is not to be confused with the GeForce GTX 660 based on the GK106 graphics Prozessores.
GK106
The GK106 GPU has been well introduced half a year after the GK104 and GK107, which is probably due to a redesign of four to five shader clusters (hence arises the special feature that the GK106 as the only Kepler GPU an asynchronous ratio of GPC to the SMX- blocks in full configuration has ). Ultimately, the GK106 from 2.54 billion transistors on a die area of 214 mm ² and has 960 stream processors and 80 texture units.
The first graphics card based on the Nvidia GK106 presented on 13 September 2012 with the Geforce GTX 660 before. The GeForce GTX 660 had a high market relevance for Nvidia because AMD had after 7850 and 7870 presents the Radeon HD already in March 2012, Nvidia could have no competitive offers for this important market segment around 200 €. Presumably, this situation led to the redesign of the GK106, as it in its original form with only four shader clusters could not keep up with the AMD competition and Nvidia had to give up this market segment than half a year. Ultimately placed Nvidia Geforce GTX 660 both the performance and the price, between the AMD competitor Radeon HD 7850 and 7870th The reviews for the GeForce GTX 660 coming in the trade press from very different, which was partly due to the fact that Nvidia no reference design pretending and thus the hardware tester completely different test samples of various board partners were given, which differed both in terms of performance, but especially during cooler design (and thus the noise ). The Geforce GTX 660 Nvidia attacked again back to the asynchronous memory configuration from 2GB Vram and a 192 bit memory interface. In contrast to the Geforce GTX 660 Ti, was observed in which some performance issues, showed no disadvantage in 3D applications on the GeForce GTX 660 by this configuration.
The second graphics card based on the GK106 presented Nvidia on 9 October 2012 with the GeForce GTX 650 Ti The naming caused this confusion, since even a Geforce GTX was introduced 650 before, but based on the much slower GK107 GPU. The GK106 comes to the GTX 650 Ti with a disabled shader clusters, therefore, making four active clusters are available. In addition, the memory interface has been reduced to 128 bits and lacked the GPU Boost feature. Nvidia placed the card against the Radeon HD 7770, against which one could have about 10 percent higher performance at a slightly lower power consumption. However, Nvidia put the starting price at too high, which caused next to the naming of criticism.
To close the relatively large performance gap between the GTX 660 and the GTX 650 Ti and to respond to the Radeon HD 7790 from AMD, Nvidia introduced on 26 March 2013, the GeForce GTX 650 Ti a boost. Contrary to the naming is at the GTX 650 Ti Boost less a GTX 650 Ti drilled, but a Geforce GTX 660, in which a SMX- block is disabled (that are still active four out of five shader cluster). The remaining specifications are with the Geforce GTX 660 is identical. The map thus reached an approximately the performance of the AMD Radeon HD competitors 7850th
GK107
The GK107 GPU is the smallest GPU Kepler generation. He has only two shader clusters and thus comes to 384 stream processors and 32 texture units. The GK107 consists of 1.3 billion transistors on a die area of 118 mm ². No previously published GK107 graphics card supports GPU Boost feature, which presently is not clear whether this feature is generally not supported by the GK107 or Nvidia has not been activated. The GK107 used Nvidia initially only in the mobile sector before April 24, 2012, the GeForce GT 630 and GT 640 were presented. These cards were provided along with some reprints of the older Fermi generation, but only for the OEM market. Only on 5 June 2012 presented Nvidia also a Geforce GT 640 for the retail market, but was heavily criticized in the press, which was due to the high price. On September 13, 2012 was followed by the improved GeForce GTX 650, in which Nvidia DDR3 replaced by faster GDDR5 memory. Despite the improvements in performance thus achieved, this card was negatively rated, because they compared to AMD competition was also too expensive.
GK208
In the GK208 GPU is in the broadest sense to create a new revision of the GK107 GPU with only a grid partition. In contrast to the rest of Kepler GPUs Nvidia was no official value or transistor die size known (the latter can be with the help of the screenshots to around 90 mm ² estimate ). The GPU was initially only installed in the mobile space before Nvidia on May 29, 2013 630 and 640 presented the second revision of the Geforce GT based on the GK208. Nvidia specifies for the GK208 Kepler GPU as the only official support for DirectX 11.1, but with the "Feature Level 11.0 ", so the exact support situation is unclear.
Rebranding
As is common already in older generations graphics cards, Nvidia repeated at the Geforce 600 series is the practice to launch the older generation graphics cards under the new name. This method, which is practiced primarily for the OEM market is called a " rebranding ". First put Nvidia on April 3, 2012, the GeForce 510 and GT 520 as GeForce 605 and GT 620 scratch. On April 24, the method with the Geforce GT 545 (DDR3) and GTX 560 SE was repeated as Geforce GT 640 and GT 645. On May 15, then again followed the Geforce GT 520 as GeForce 610 and GeForce GT 430, GT 440 ( DDR3) and GT 440 ( GDDR5 ) as Geforce GT 620, GT 630 (DDR3) and GT 630 ( GDDR5 ).
Architecture
The basis of the Geforce 600 series, the newly developed Kepler architecture, named after the German mathematician Johannes Kepler, which replaces the existing Fermi architecture is used. Although Nvidia called the Kepler architecture as a new development, this is actually an evolution of the previous generation Fermi dar. Primary change represents the elimination of the "hot clock " for the shader units and stream processors dar. 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. Thus, Nvidia was able to achieve higher performance with less shader units, however, it also meant that the GPU had a relation to the GPUs worse from competitor AMD energy efficiency. With the removal of the "hot clock " Nvidia could fix this shortcoming and simultaneously block more shader units, as these are considered imaginary bill is now less space on the need (Nvidia that at a 10 percent lower power consumption 80 percent more die area for to use the shader units).
Regardless of the revised stream processors Nvidia has simplified the " scheduling ", the calculation commands and more outsourced to the software. In addition, the raster engine is now capable of supplying all ROPs per cycle with a pixel which has not been possible on the Fermi generation.
The basic structure of the Kepler architecture is similar to Fermi: As before, the GPUs from so-called " Graphics Processing Clusters" short GPC, together ( the GK104 is an example of four GPCs ) set. Each GPC consists of two shader clusters ( from Nvidia called SMX- blocks), turn accommodate each 192 stream processors, as well as load and store units and special function units. Furthermore, 16 instead of four to eight texture units installed, which can address a pixel per clock and texture in each shader cluster now. The structure of the grid units unchanged: As before, this is divided into ROP clusters, wherein each cluster of eight " raster operation Processors " and a 64-bit memory controller concludes.
A special feature of the Kepler architecture, the GPU Boost feature dar. It is a dynamic overclocking function that always clocked the GPU higher if a particular fixed set of Nvidia "Power Target Limit ' is not reached. Other factors include the power, temperature, and the accompanying currents. Graphics memory is not affected by the function.