NVIDIA has been involved in the detailed development of architecture for a sufficiently long time to maintain decent competitiveness of its products. Thus, in the overwhelming majority of cases, the company spends approximately two years moving from one series to the next.
So the release of GTX 470 video cards was delayed, as the developers were forced to rework the architecture quite strongly, there were some problems with production, in the end, the created chip turned out to be too hot during its work. Thus, the cycle remained almost exactly the same, and almost two years have passed since the previous card was released. However, one way or another, the released GTX 470 cards initially claimed to be called the technological leader of their time. There is no doubt that they were.
GeForce 400
By tradition, NVIDIA provided at the same time two models of video cards, GTX 470 and 480, however, it is the first version that is more common among users today.
Experts immediately turned their attention to the number of transistors that was used in Fermi. According to this parameter, it was possible to indirectly say how good the new chip is, because it was twice as productive compared to the previous flagship of this company, and also one and a half times compared to the more modern Cypress, which is very very good result.
Of course, this indicator directly affected the heating. In the process of creating the GTX 470, it was decided to use a 40-nm technical process, as a result of which the difference in the thermal package with similar devices is approximately the same as the difference in the number of transistors.
Technical features
Initially, it was decided to use 512 CUDA processors, but over time this decision was revised, as a result of which the figure was slightly reduced to reduce heat generation, as well as increase the output of really working chips. Nevertheless, in the end, the cards were still quite expensive for the vast majority of users.
A strange fact was that the memory frequency of even the most productive model is lower compared to the competing HD 5850 solution . The total video bandwidth of the GeForce GTX 470 is higher, but due to the fact that it used faster chips, there was the possibility of further overclocking .
How does she behave in games?
Both models provide support for DirectX 11. After the video cards of the Evergreen model began to appear on the market, there was practically nothing to test the capabilities of this device, but the situation remained the same with the release of the GeForce GTX 470. Games that could support the new API were isolated cases, but over time, of course, a large number of different projects were released for which such a video card was needed. Even today, its use is optimal for the predominant majority of games, and if it is impossible to play the most modern on the maximum requirements, then it will maintain medium or at least minimal ones stably.
As usual, the company decided to bet in favor of the future, optimizing its NVIDIA GeForce GTX 470 to work with the capabilities offered by the latest DirectX, as well as paying special attention to GPGPU.
Architecture of the new chip
The new device used a chip with a new architecture called the GF100. At first glance, there were no significant differences from its predecessor, while the design of the new product remained scalar, as a result of which there were certain limitations. The composition of universal processors has slightly changed, which it was decided to combine into fairly large blocks. In the GT200 model, such blocks were called TPC, and each of them included all the necessary components that may be needed when working with graphics, excluding rasterization blocks. In the future, the company quickly began to produce a lot of cheapened versions of the NVIDIA GeForce GTX 470, in which such blocks were either disabled or modified.
In each such block of the previous model of the chip there were three SM (streaming multiprocessors), as well as eight texture units and a cache of the first level with a capacity of 16 Kb. In the block of the new device, in turn, one Raster Engine was used, as well as four SMs, each of which had its own cache memory, as well as texture blocks. In the chip itself, there are four blocks at the same time, in total providing the user with 32 pixels per clock.
Streaming multiprocessors
If we consider devices at the SM level, the difference becomes quite significant, since when it was released from the NVIDIA GTX 470, many experts drew attention to the fact that here SMs are much larger and independent. In the last generation of cards, there were 8 SPs, as well as 2 SFUs, but now their number and ratio have been changed, as a result of which each device has 4 SFUs, 32 CUDA Core, as well as 16 Load / Store units.
In each of these third-generation multiprocessors, there were several texture processors, as well as a shared memory that provided hardware tessellation.
Due to the fact that the texture processors, as well as all the corresponding cache memory, were moved inside the multiprocessors, the developers were able to achieve much more efficient operation of these devices. It was also decided to radically redesign the internal architecture of these devices, increase their frequency of operation, plus triple increase the amount of second-level memory that is actively used in the texturing process.
Common memory
The shared memory was combined with L1 cache, as a result of which the corresponding amount of memory began to be allocated correctly depending on what kind of task the device is currently performing. Previously, the total memory included only 16 KB, but at the moment, depending on the type of task being performed, the device will allocate 64 KB of memory. If at the moment a large number of physical calculations are carried out, then in this case the GTX 470 graphics card distributes 48 Kb of this memory to L1. If various processes refer to certain information a large number of times, then it is stored in shared memory. Among other things, useful for GPGPU is also the tripled L2 cache size in the GTX 470. The device characteristics, therefore, more and more went ahead compared to competitors.
Tessellation
Tessellation is one of the key features of the Radeon HD 5000 card . The NVIDIA GeForce GTX 470 graphics card , in turn, is better sharpened for implementing this algorithm.
The DirectX 11 API supports certain “displacement maps,” which are volume textures. Thus, each pixel in the picture of this texture will determine its height. This texture is superimposed in the game in order to make the image more detailed.
This is where the tessellation that the GTX 470 is equipped with comes in. A review of this algorithm showed that a whole mosaic is formed from one polygon. This technology is used regardless of displacement maps, but the most impressive effect can be achieved only if they are used together. The vertices of the formed polygons will fall or rise to a certain height, bringing the system closer to the model, which is set by the room map.
What are the benefits?
In fact, the model turns out to be absolutely identical to the standard high-poly, while at the same time it is considered low-poly, because it will give the correct flare and correctly cast a shadow, while at the same time spending an order of magnitude less resources.
In other words, the NVIDIA GTX 470 features are such that they allow you to quickly load models into the memory of the card, while subsequently you will not need to load and unload objects with different levels of detail, which was relevant for devices of the past generation. All recounting processes are carried out directly inside the card during various procedures, which provides much greater scalability. Depending on the presence of completely free computing power, it will be possible to approach the displacement map with a certain accuracy, which is quite important in terms of the geometric performance of the GTX 470. User reviews say that this technology significantly improves the overall performance of the computer during complex graphics processes.
What do users say?
Feedback from users updating their PC by installing the described card indicates that after the upgrade they managed to set the maximum settings in all games. At the same time, the game did not freeze, no brakes were observed.
Of course, such reviews are typical for 2010-2012, at the moment the card no longer provides its owner with such unlimited possibilities. Nevertheless, during this time technology has stepped far forward, and there are already such games that you can play with such a video card only at the minimum settings.
3D vision
3D technology today has become extremely widespread, from televisions to monitors. Even AMD has been releasing its own glasses for a long time, as well as the corresponding software that allows the manufacturer's video cards to display high-quality stereoscopic images.
In the GeForce GTX 470, the specifications also provide for the possibility of using this technology, and despite the fact that this significantly affects performance, the overall experience of the gameplay or watching a movie becomes completely different.
The entire set of tools provided by various graphics cards and APIs is quite difficult to master for ordinary users and is considered absolutely useless if no one uses it. It is for this reason that NVIDIA created a separate unit that works with modern game developers. Of course, attention was always paid to the most successful projects, and mutually beneficial cooperation was established, because in the end the game itself became much faster and more beautiful if it was launched by a user on a computer with this card.
Cooling system
To ensure proper cooling of the card, the company developed a specialized cooler, which is one of the most powerful reference models among analogues. The design of the device is made in the form of a standard turbine, but at the same time it has several features that definitely deserve special attention.
The basis of the system is a mechanical frame with a plastic casing fixed to it. Through specialized thermal pads, the heat generated during operation of the memory chips and the subsystem is transferred to the frame. From the GF 100 itself, heat is removed using five special heat pipes in direct contact with the chip.
Direct contact of the heat pipes of the chip is often used in processor cooling systems, and such a system cannot be called obviously more effective than the standard one. But in this case, it means that when the GeForce GTX 470 was being developed, the developers paid special attention to the cooling system.