The intensity of competition between ATI and NVIDIA (namely, they act as "trendsetters" in the market of graphics processors) and, accordingly, the rate of development of the industry are extremely high: a new generation of graphics card replaces the previous average – once a year and a half. So many of those who do not have regular opportunities to keep track of events currently on the market of computer graphics, are confident that there is a real chaos reigns, and to understand the situation, "mere mortals" can not be. In part they are right now in the shops can be found over a hundred different implementations of graphics produced by the most bizarre combinations of different types of graphics chips and memory, and their operating frequencies and bit tires. But is it all bad as it seems at first sight? Let's try to figure it out.
NVIDIA GPUs line NV4x
The most widespread family of graphics processors by NVIDIA at the moment, is a series of NV4x. It is because of this very successful series of chips, the California giant has managed not only to rectify its position in the market, undermined by the failure of the previous line NV3x, but also to demonstrate to the world sustainable growth trend, particularly noticeable on the background of certain failures of ATI in the past few months.
The flagship model line is NV4x chip NV40, saw the light in April 2004. It is well-developed, but rapidly aging 130-nm process in the factories IBM, its 222 million transistors consume up to 120 watts of energy, so the "power" features regular AGP 8x interface is sorely lacking and the cards made based on NV40, usually set of two (!) of additional power connector. And the cooling system is all this "oven" should be very weak. Along with the NV40, produced and its PCI Express-modification of the NV45, the only difference is the base model is integrated into the body of the chip AGP-PCI-E bridge HSI. In the NVIDIA product line meets another, and the chip NV48, which is different from the base NV40 only that is produced in factories TSMC. Thus, NVIDIA entirely refused the services of IBM in the manufacture of chips, and returned to his old, and apparently more acceptable than IBM, a technology partner.
256-bit quad (organization 64h4) memory interface NV40 provides connectivity up to 1GB of memory of any kind – as usual DDR1/DDR2, and specially designed for use with video devices GDDR3.
As for the architectural features of NV40, it is worth noting that it was the first graphics processor, which features a fully compliant DirectX 9.0, in other words, its computational capabilities allow you to shaders SM (Shader Model) 3.0. Provide a high performance 6 vertex and 16 pixel pipelines (each of which, in turn, has two shader and texture in one piece). At the same time can operate either two shader units, or a texture and a shader. Thus, NV40 enables you to cycle up to 16 texture operations and 32 operations, with the depth and stencil buffer, providing anisotropic filtering with an aspect ratio up to 16:1 inclusive.
On the basis of the core NV40/NV45/NV48 issued cards: GeForce 6800 Ultra, GeForce 6800 GT, GeForce 6800 and GeForce 6800 LE. Typical values of a core clock / memory bus up the GeForce 6800 Ultra 400 MHz / 1.1 GHz, a modification of the GeForce 6800 GT, they are reduced to 350 MHz / 1 GHz processor, the GeForce 6800 – up to 325 MHz MGts/700, while the GeForce 6800 LE frequency parameters are not specified at all – all given at the discretion of manufacturers of graphics cards. In doing so, the latter two also reduced to 12 the number of pixel pipelines. Moreover, in contrast to common practice, when the "cuts" made by the software off a certain number of processors (which enables the people, "craftsmen" by a straightforward operation to connect the unused blocks, having thus a full-featured GeForce 6800 for little money), in this case NVIDIA has released a special "truncated" core NV41. It has only 12 "physical" pixel processors, while everything else is fully consistent with the underlying NV45. The core of NV42 is a 110-nm version of the NV41. By the way, it is the basis of the latest news NVIDIA – GeForce 6800 GS (the frequency of the chip – 425 MHz Memory – 1000 MHz), designed to compete with ATI Radeon X1600 XT.
For mid-range graphics cards, NVIDIA in August 2004 released a graphics engine NV43, which is, however, somewhat simplistic (by reducing the number of vertex and pixel processors and memory controller channels) solution based on the architecture of the NV40. However, the NV43 performed on 0.11-micron TSMC (the number of transistors in the core is 146 million), and, besides, he became the first NVIDIA graphics core with integrated controller, PCI Express (available Broadcast Interface PCI-E in the APG 8X with two-way PCI-E-AGP bridge HSI). Thanks to the "finer" process technology and fewer actuators in the nucleus, heat NV43 does not exceed 70 watts, that is, the PCI-Express card slot for additional power supply is not needed.
A set of computational resources NV43 exactly twice smaller than the older model NV40 – he has eight pixel pipelines and three vertex, and the memory bus width is reduced from 256 to 128 bits.
At the present time is represented by two Mainstream-modifying graphics cards based on GPU NV43 – GeForce 6600 and GeForce 6600 GT, as well as a class Low-End – GeForce 6200 (though it has recently been renamed the GeForce 6600 LE). Core and memory bus GeForce 6600 are 300 and 550 (sometimes – 500) MHz, and a modified GT – 500 and 1000 MHz respectively. In the GeForce 6200/6600 LE uses "truncated" core NV43V with four (instead of 
pixel pipelines, and its frequency range corresponds to the GeForce 6600.
Thanks to the high clock frequency of the chip, although only 8 pixel rendering pipelines, NV43 provides even more fillrate than 12 pipeline GeForce 6800. However, realistically achieve a level of performance of his "senior fellow" GeForce 6600 GT is not given – affects the use of 128-bit memory bus and a decrease of up to 3 vertex processors, although it is significantly cheaper and allowed the 6600-S series. Therefore, it is not surprising that the video card GeForce 6600 GT chips and, especially, GeForce 6600, are now showing the best correlation "price / performance."
In the low-end model of the family of graphics processors NVIDIA GeForce 6×00 – GeForce 6200TS is used primarily by the core NV44, made of 0.11-micron design rules. As in the case of NV43, PCI Express his native (that is implemented on a chip) and AGP 8x version of the chip is referred to as NV44A. NV44 has 3 vertex processors, as in NV43, and 4 pixel, any global architectural differences from NV40 and NV43.
The main performance limiter for NV44, a memory subsystem – the width of tires is limited to 64 bits. The index denotes the TC support technology TurboCache, designed to engage a portion of system memory for frame buffer rendering, good bandwidth PCI-E x16 for this lack. Of course, these cards lose much of its full performance level counterparts, but little benefit from them at cost. And for budget decisions, every dollar saved can be decisive in promoting solutions to market one or another of competing firms. However, for someone who wants to save money by purchasing a card, such gains will be illusory. After all, the system memory, especially DDR2, from which "ottyapyvaet" its part of the video system, is not dimensionless, and is not free.
NVIDIA GPUs line G7x
Mid-summer 2005 was a milestone in the history of NVIDIA – was released on next-generation graphics processor G70. But despite the new code name for the chip, the architecture can not be considered entirely new – it is the next stage of evolution familiar to us all of the family architecture NV4x (as evidenced by the original code word chip – NV47).
Plus to this, NVIDIA has not released its risk and well-developed 110-nm TSMC, it is not surprising that the family of graphics cards GF7800GTX became available in large quantities immediately after the announcement.
The main innovation of G70, all based on the same set of Shaders Shader Model 3.0 (SM 3.0), was increased to 24 (or, in the terminology of NVIDIA, 6 quad processors) of a slightly improved compared to NV4x, pixel processors and vertex – 8, the amount of addressable memory type GDDR3 can reach 1 GB. In addition, the optimization was carried out food and energy, so that typical power consumption of GeForce 7800 GTX does not exceed 110 watts, it has remained at the level of the GeForce 6800 Ultra, despite a 30% increase in the number of transistors.
In the G70 hardware support appeared a number of potential "hits" in the near future: video playback in HDTV, as well as support for the important special features graphics driver model Windows Vista.
In addition to High-End video card GeForce 7800 GTX (430/1200 MHz) processor and the G70 is set in its little brother of a more modest – GeForce 7800 GT, GTX different from not only reduced (up to 400/1000 MHz) frequencies, but also reduce the number of pipelines (20 pixel and 7 vertex) rendering, which ensures him a place (in terms of performance) somewhere between the GeForce 7800 GTX and GeForce 6800 Ultra.
Successive new chip NVIDIA G72 and should be a G74 (the official name of the GeForce 7600 and GeForce 7200), which is expected in February and March 2006. They are likely to be truncated versions of the base G70, and will be the first graphics chip NVIDIA, let out on the 0.09 micron technology. They should be replaced in the NV43 and NV44 Mainstream-and Low-End sector, respectively. Thus, in the next six months, NVIDIA is going to completely replace all the chips on the news from NV4x line G7x. In addition, NVIDIA in early 2006, plans to introduce chip G71, which is a 0.09 micron version of the G70 with a smaller die size and low power consumption.