How to track GPU usage in Windows Task Manager. Built-in and integrated graphics in the processor What is copy in the GPU parameters

The graphics processing unit (GPU) is no less important component of the SoC of a mobile device than the CPU. Over the past five years, the rapid development of the Android and iOS mobile platforms has spurred mobile GPU developers, and today no one will be surprised by mobile games with PlayStation 2 level or even higher 3D graphics. I dedicated the second article in the “Educational program on mobile hardware” series to graphic processors.

Currently, most graphics chips are produced using cores: PowerVR (Imagination Technologies), Mali (ARM), Adreno (Qualcomm, formerly ATI Imageon) and GeForce ULP (nVIDIA).

PowerVR is a division of Imagination Technologies, which in the recent past developed graphics for desktop systems, but was forced to leave this market under pressure from ATI and nVIDIA. Today, PowerVR develops perhaps the most powerful GPUs for mobile devices. PowerVR chips are used in the production of processors by companies such as Samsung, Apple, Texas Instruments, etc. For example, different GPU revisions from PowerVR are installed in all generations of the Apple iPhone. The 5 and 5XT series of chips remain relevant. The fifth series includes single-core chips: SGX520, SGX530, SGX531, SGX535, SGX540 and SGX545. 5XT series chips can have from 1 to 16 cores: SGX543, SGX544, SGX554. The specifications of the 6 series (Rogue) are still being finalized, but the performance range of the series chips is already known – 100-1000GFLOPS.

Mali are GPUs developed and licensed by UK-based ARM. Mali chips are an integral part of various SoCs manufactured by Samsung, ST-Ericsson, Rockchip, etc. For example, Mali-400 MP is part of the Samsung Exynos 421x SoC used in smartphones such as the Samsung Galaxy SII and SIII, two generations of “smartphone tablets” ? Samsung Note. Current today is the Mali-400 MP in dual- and quad-core versions. The Mali-T604 and Mali-T658 chips are coming, the performance of which is up to 5 times higher than that of the Mali-400.

Adreno are graphics chips that are developed by the eponymous division of the American Qualcomm. The name Adreno is an anagram of Radeon. Before Qualcomm, the division belonged to ATI, and the chips were called Imageon. For the last few years, Qualcomm has been using 2xx series chips in the production of SoCs: Adreno 200, Adreno 205, Adreno 220, Adreno 225. The last one on the list is a completely fresh chip - made using 28nm technology, the most powerful of the Adreno 2xx series. Its performance is 6 times higher than that of the “old man” Adreno 200. In 2013, more and more devices will receive Adreno 305 and Adreno 320 graphics processors. Already now the 320th is installed in the Nexus 4 and the Chinese version of Nokia Lumia 920T, according to some parameters of the chip 2 times more powerful than the 225th.

GeForce ULP (ultra-low power) is a mobile version of the video chip from nVIDIA, part of the Tegra system-on-chip of all generations. One of Tegra's most important competitive advantages is specialized content intended only for devices based on this SoC. nVIDIA has traditionally had close relationships with game developers, and their Content Development team works with them to optimize games for GeForce graphics solutions. To access such games, nVIDIA even launched the Tegra Zone Android application, a specialized analogue of the Android Market, where you can download applications optimized for Tegra.

GPU performance is typically measured in three ways:

– the number of triangles per second, usually in millions – Mega (MTriangles/s);

– number of pixels per second, usually in millions – Mega (MPixel/s);

– the number of floating point operations per second, usually in billions – Giga (GFLOPS).

“Flops” requires a little explanation. FLOPS (FLoating-point Operations Per Second) is the number of computational operations or instructions performed on floating-point operands per second. A floating point operand is a non-integer number (it would be more correct to say “floating point”, since the symbol that separates the integer part of a number from the fractional part in Russian is the comma). It will help you understand which graphics processor is installed in your smartphone. ctrl+F and the table below. Please note that the GPUs of different smartphones operate at different frequencies. To calculate the GFLOPS performance for a specific model, you need to divide the number indicated in the “GFLOPS performance” column by 200 and multiply by the frequency of the individual GPU (for example, in the Galaxy SIII, the GPU operates at a frequency of 533 MHz, which means 7.2 / 200 * 533 = 19.188) :

Name of smartphone/tablet	CPU	GPU	Performance in GFLOPS
Samsung Galaxy S 4	Samsung Exynos 5410	PowerVR SGX544MP3	21.6 @200MHz
HTC One	Qualcomm Snapdragon 600 APQ8064T	Adreno 320	20.5 @200MHz
Samsung Galaxy S III, Galaxy Note II, Galaxy Note 10.1	Samsung Exynos 4412	Mali-400 MP4	7.2 @200MHz
Samsung Chromebook XE303C12, Nexus 10	Samsung Exynos 5250	Mali-T604 MP4	36 @200MHz
Samsung Galaxy S II, Galaxy Note, Tab 7.7, Galaxy Tab 7 Plus	Samsung Exynos 4210	Mali-400 MP4	7.2 @200MHz
Samsung Galaxy S, Wave, Wave II, Nexus S, Galaxy Tab, Meizu M9	Samsung Exynos 3110	PowerVR SGX540	3.2 @200MHz
Apple iPhone 3GS, iPod touch 3gen	Samsung S5PC100	PowerVR SGX535	1.6 @200MHz
LG Optimus G, Nexus 4, Sony Xperia Z	Qualcomm APQ8064(Krait cores)	Adreno 320	20.5 @200MHz
HTC One XL, Nokia Lumia 920, Lumia 820, Motorola RAZR HD, Razr M, Sony Xperia V	Qualcomm MSM8960(Krait cores)	Adreno 225	12.8 @200MHz
HTC One S, Windows Phone 8x, Sony Xperia TX/T	Qualcomm MSM8260A	Adreno 220	~8.5* @200MHz
HTC Desire S, Incredible S, Desire HD, Sony Ericsson Xperia Arc, Nokia Lumia 800, Lumia 710	Qualcomm MSM8255	Adreno 205	~4.3* @200MHz
Nokia Lumia 610 LG P500	Qualcomm MSM7227A	Adreno 200	~1.4* @128MHz
Motorola Milestone, Samsung i8910, Nokia N900	TI OMAP3430	PowerVR SGX530	1.6 @200MHz
Samsung Galaxy Nexus Huawei Ascend P1, Ascend D1, Amazon Kindle Fire HD 7″	TI OMAP4460	PowerVR SGX540	3.2 @200MHz
RIM BlackBerry Playbook, LG Optimus 3D P920, Motorola ATRIX 2, Milestone 3, RAZR, Amazon Kindle Fire first and second generations	TI OMAP4430	PowerVR SGX540	3.2 @200MHz
Motorola Defy, Milestone 2, Cliq 2, Defy+, Droid X, Nokia N9, N950, LG Optimus Black, Samsung Galaxy S scLCD	TI OMAP3630	PowerVR SGX530	1.6 @200MHz
Acer Iconia Tab A210/A211/A700/A701/A510, ASUS Transformer Pad, Google Nexus 7, Eee Pad Transformer Prime, Transformer Pad Infinity, Microsoft Surface Sony Xperia Tablet S, HTC One X/X+, LG Optimus 4X HD, Lenovo IdeaPad Yoga	nVidia Tegra 3	GeForce ULP	4.8 @200MHz
Acer Iconia Tab A500, Iconia Tab A501, Iconia Tab A100, ASUS Eee Pad Slider, Eee Pad Transformer, HTC Sensatoin/XE/XL/4G, Lenovo IdeaPad K1, ThinkPad Tablet, LG Optimus Pad, Optimus 2X, Motorola Atrix 4G, Electrify, Photon 4G, Xoom, Samsung Galaxy Tab 10.1, Galaxy Tab 8.9, Sony Tablet P, Tablet S	nVidia Tegra 2	GeForce ULP	3.2 @200MHz
Apple iPhone 5	Apple A6	PowerVR SGX543MP3	19.2 @200MHz
Apple iPad 2, iPhone 4S, iPod touch 5gen, iPad mini	Apple A5	PowerVR SGX543MP2	12.8 @200MHz
Apple iPad, iPhone 4, iPod touch 4gen	Apple A4	PowerVR SGX535	1.6 @200MHz

* – data is approximate.

Let me give you another table with the absolute performance values ​​of the most popular smartphones in the upper price range:

* – unofficial data.

The power of mobile graphics is growing from year to year. Already this year we can see PS3/X-Box360 level games in top smartphones. At the same time as the power, the power consumption of the SoC increases significantly and the autonomy of mobile devices decreases indecently. Well, let's wait for a breakthrough in the production of power supplies!

Another energy hog in a modern mobile device is, of course, the display. Mobile phone screens are becoming more and more beautiful. The displays of smartphones released just a year apart differ dramatically in picture quality. In the next article in the series, I will talk about displays: what types they are, what PPI is, what power consumption depends on, and so on.

The source material for rendering is a set of triangles of various sizes, which make up all the objects of the virtual world: landscape, game characters, monsters, weapons, etc. However, the models themselves, created from triangles, look like wire frames. Therefore, textures are superimposed on them - colored two-dimensional “wallpapers”. Both textures and models are placed in the memory of the graphics card, and then, when creating each frame of game action, a rendering cycle is performed, consisting of several stages.

1. The game program sends information to the graphics processor that describes the game scene: the composition of the objects present, their color, position relative to the observation point, lighting and visibility. Additional data is also transmitted that characterizes the scene and allows the video card to increase the realism of the resulting image by adding fog, blur, glare, etc.

2. The graphics processor places three-dimensional models in the frame, determines which of the triangles included in them are visible and cuts off those hidden by other objects or, for example, shadows.

Then light sources are created and their effect on the color of the illuminated objects is determined. This stage of rendering is called “transformation and lighting” (T&L - Transformation & Lighting).

3. Textures are applied to visible triangles using various filtering technologies. Bilinear filtering involves overlaying two versions of a texture with different resolutions on a triangle. The result of its use is clearly visible boundaries between areas of clear and blurred textures that appear on three-dimensional surfaces perpendicular to the viewing direction. Trilinear filtering, using three variations of the same texture, allows you to create softer transitions.

However, as a result of using both technologies, only those textures that are located perpendicular to the axis of vision look truly clear. When viewed from an angle they become very blurry. To prevent this, anisotropic filtering is used.

This texture filtering method is set in the video adapter driver settings or directly in the computer game. In addition, you can change the strength of anisotropic filtering: 2x, 4x, 8x or 16x - the more “X’s”, the clearer the images on inclined surfaces will be. But as the filtering strength increases, the load on the video card increases, which can lead to a decrease in operating speed and a decrease in the number of frames generated per unit of time.

Various additional effects can be used at the texturing stage. For example, Environmental Mapping allows you to create surfaces in which the game scene will be reflected: mirrors, shiny metal objects, etc. Another impressive effect is achieved with the use of bump mapping, thanks to which light falling on a surface at an angle creates the appearance of a relief.
Texturing is the last stage of rendering, after which the image enters the frame buffer of the video card and is displayed on the monitor screen.

Electronic components of the video card

Now that it has become clear how the process of constructing a three-dimensional image occurs, we can list the technical characteristics of the video card components that determine the speed of the process. The main components of a video card are the graphics processor (GPU - Graphics Processing Unit) and video memory.

GPU

One of the main characteristics of this component (as well as the PC central processor) is the clock speed. All other things being equal, the higher it is, the faster data processing occurs, and therefore the number of frames per second (FPS - frames per second) in computer games increases. The frequency of the graphics processor is an important, but not the only parameter that affects its performance - modern models produced by Nvidia and ATI, which have a comparable level of performance, are characterized by different GPU frequencies.

High-performance Nvidia adapters feature GPU clock speeds from 550 MHz to 675 MHz. Mid-range and cheap low-performance cards have a GPU operating frequency of less than 500 MHz.
At the same time, the GPUs of “top” ATI cards have frequencies from 600 to 800 MHz, and even the cheapest video adapters have a GPU frequency that does not drop below 500 MHz.

However, even though Nvidia GPUs are clocked lower than ATI-designed GPUs, they provide at least the same level of performance, and often better. The fact is that other GPU characteristics are no less important than the clock speed.

1. The number of texture modules (TMU - Texture Mapping Units) - GPU elements that perform texture mapping on triangles. The speed of building a 3D scene directly depends on the number of TMUs.
2. The number of rendering pipelines (ROP - Render Output Pipeline) - blocks that perform “service” functions (a couple of examples, pls). Modern GPUs tend to have fewer ROPs than texture units, and this limits overall texturing speed. For example, the Nvidia GeForce 8800 GTX video card chip has 32 “texture units” and 24 ROPs. The processor of the ATI Radeon HD 3870 video card has only 16 texture models and 16 ROPs.

The performance of texture modules is expressed in a value called fillrate - texturing speed, measured in texels per second. The GeForce 8800 GTX video card has a fill rate of 18.4 billion tex/s. But a more objective indicator is the fill rate, measured in pixels, since it reflects the speed of the ROP. For the GeForce 8800 GTX this value is 13.8 billion pixels/s.
3. The number of shader units (shader processors), which - as the name suggests - are responsible for processing pixel and vertex shaders. Modern games use shaders heavily, so the number of shader units is critical to determining performance.

Not long ago, GPUs had separate modules for pixel and vertex shaders. Nvidia GeForce 8000 series video cards and ATI Radeon HD 2000 adapters were the first to switch to a unified shader architecture. The graphics processors of these cards have units capable of processing both pixel and vertex shaders - universal shader processors (stream processors). This approach allows you to fully utilize the computing resources of the chip for any ratio of pixel and vertex calculations in the game code. In addition, in modern GPUs, shader units often operate at a frequency higher than the GPU clock frequency (for example, the GeForce 8800 GTX has this frequency of 1350 MHz versus the “general” 575 MHz).

Please note that Nvidia and ATI count the number of shader processors in their chips differently. For example, the Radeon HD 3870 has 320 such units, and the GeForce 8800 GTX has only 128. In fact, ATI indicates their component components instead of entire shader processors. Each shader processor contains five components, so the total number of shader units on the Radeon HD 3870 is only 64, which is why this video card runs slower than the GeForce 8800 GTX.

Video card memory

Video memory in relation to the GPU performs the same functions as RAM does in relation to the PC's central processor: it stores all the “building material” necessary to create an image - textures, geometric data, shader programs, etc.

What video memory characteristics affect graphics card performance?

1. Volume. Modern games use a huge amount of high-resolution textures, and they require a corresponding amount of video memory to accommodate them. The bulk of “top” video adapters and mid-price cards produced today are equipped with 512 MB of memory, which cannot be increased later. Cheaper video cards are equipped with half the amount of memory, which is no longer enough for modern games.

If there is insufficient memory, the GPU is forced to constantly load textures from the PC's RAM, communication with which is much slower, and as a result, performance can noticeably decrease. On the other hand, an excessively large amount of memory may not provide any speed increase, since the additional "space" simply will not be used. Buying a video adapter with 1 GB of memory only makes sense if it belongs to the “top” products (ATI Radeon HD 4870, Nvidia GeForce 9800 video cards, as well as the latest GeForce GTX 200 series cards).

2. Frequency. This parameter for modern video cards can vary from 800 to 3200 MHz and depends, first of all, on the type of memory chips used. DDR 2 chips can provide operating frequencies up to 800 MHz and are used only in the cheapest graphics adapters. GDDR 3 and GDDR 4 memory increases the frequency range up to 2400 MHz. The latest ATI Radeon HD 4870 graphics cards use GDDR-5 memory with a fantastic frequency of 3200 MHz.

Memory frequency, like GPU frequency, has a big impact on video card performance in games, especially when using full-screen anti-aliasing. All other things being equal, the higher the memory frequency, the higher the performance, because The GPU will be idle less while waiting for data to arrive. Memory speeds of 1800 MHz are the lower limit separating high-end cards from slower ones.

3. The video memory bus width has a much stronger effect on the overall performance of the card than the memory frequency. It shows how much data the memory can transfer in one clock cycle. Accordingly, doubling the memory bus width is equivalent to doubling its clock frequency. The bulk of modern video cards have a 256-bit memory bus. Reducing the bit depth to 128 or, even more so, to 64 bits causes a strong blow to performance. On the other hand, in the most expensive video cards the bus can be “expanded” to 512 bits (so far only the latest GeForce GTX 280 can boast of this), which turns out to be very useful, taking into account the power of their graphics processors.

Where to find information about the technical characteristics of a video card

If a graphics card has some outstanding parameters (high clock speed of the processor and memory, its capacity), then they are usually indicated directly on the box. But the most complete specifications of video adapters and the GPUs on which they are based can only be found on the Internet. General information is posted on the corporate websites of GPU manufacturers: Nvidia (www.nvidia.ru) and ATI (www.ati.amd.com/ru). Details can be found on unofficial websites dedicated to video cards - www.nvworld.ru and www.radeon.ru. The electronic encyclopedia Wikipedia (www.ru.wikipedia.org) will be a good help. Users who buy a card with an eye toward overclocking can use the resource www.overclockers.ru.

Simultaneous use of two video cards

In order to get maximum performance, you can install two video cards in your computer at once. Manufacturers have provided the appropriate technologies for this - SLI (Scalable Link Interface, used by Nvidia cards) and CrossFire (developed by ATI). In order to use them, the motherboard must not only have two PCI-E slots for video cards, but also support one of the named technologies. Many motherboards based on Intel chipsets can use ATI boards in CrossFire mode, but only boards based on chipsets from the same company can combine two (or even three!) video cards from Nvidia into one “harness.” If the motherboard does not support these technologies, two video cards will be able to work with it, but only one will be used in games, and the second will only provide the ability to display images on a pair of additional monitors.
Note that using two video cards does not double the performance. The average result you can expect is a 50% increase in speed. In addition, the full potential of the tandem will be revealed only when using a powerful central processor and a high-resolution monitor.

What are shaders

Shaders are microprograms present in the game code that can be used to change the process of constructing a virtual scene, opening up possibilities that are unattainable using traditional 3D rendering tools. Modern game graphics without shaders are unthinkable.

Vertex shaders change the geometry of three-dimensional objects, making it possible to realize natural animation of complex game character models, physically correct deformation of objects, or real waves on water. Pixel shaders are used to change the color of pixels and allow you to create effects such as realistic circles and ripples in water, complex lighting and surface relief. In addition, with the help of pixel shaders, post-processing of the frame is carried out: all sorts of “cinematic” effects of blurring moving objects, super-bright light, etc.

There are several versions of the implementation of the shader model (Shader Model). All modern video cards support pixel and vertex shaders version 4.0, which provide higher realism of effects compared to the previous third version. Shader Model 4.0 is supported by the DirectX 10 API, which runs exclusively on Windows Vista. In addition, the computer games themselves must be designed for DirectX 10.

Does an old system need an AGP video card?

If your PC's motherboard is equipped with an AGP port, the options for upgrading the video card are very limited. The maximum that the owner of such a system can afford is a Radeon HD 3850 series video card from AMD (ATI).

By modern standards, they have below average performance. Additionally, the vast majority of AGP-capable motherboards are designed for older Intel Pentium 4 and AMD Athlon XP processors, so overall system performance will still not be fast enough for modern 3D graphics. Only motherboards for AMD Ahtlon 64 processors with Socket 939 should install new video cards with an AGP port. In all other cases, it is better to buy a new computer with a PCI-E interface, DDR 2 (or DDR 3) memory and a modern CPU.

Material tags: graphic card, video, card, accelerator, graphics

Modern video cards, due to the enormous computing power they require when working with graphics, are equipped with their own commandcenter, in other words - the graphics processor.

This was done in order to “unload” the central processor, which, due to its wide “scope of application,” is simply not able to cope with the requirements that moderngame industry.

Graphics processing units (GPUs) are absolutely not inferior to central processors in complexity, but due to their narrow specialization, they are able to more effectively cope with the task of processing graphics, constructing an image, and then displaying it on the monitor.

If we talk about the parameters, they are very similar for GPUs to central processors. These are parameters already known to everyone, such as processor microarchitecture, clock frequency core work, production process. But they also have quite specific characteristics. For example, an important characteristic of a GPU is the number of pixel pipelines. This characteristic determines the number of pixels processed per GPU clock cycle. The number of these pipelines may vary, for example, in Radeon HD 6000 series graphics chips, their number can reach 96.

The pixel pipeline is engaged in calculating each subsequent pixel of the next image, taking into account its features. To speed up the rendering process, several parallel running pipelines are used that calculate different pixels of the same image.

Also, the number of pixel pipelines affects an important parameter - the filling speed of the video card. The fill rate of a video card can be calculated by multiplying the core frequency by the number of pipelines.

Let's calculate the fill rate, for example, for an AMD Radeon HD 6990 video card (Fig.2) The GPU core frequency of this chip is 830 MHz, and the number of pixel pipelines is 96. With simple mathematical calculations (830x96), we come to the conclusion that the fill rate will be equal to 57.2 Gpixel/s.

Rice. 2

In addition to pixel pipelines, there are also so-called texture units in each pipeline. The more texture units, the more textures can be applied in one pass of the pipeline, which also affects the overall performance of the entire video system. In the aforementioned AMD Radeon HD 6990 chip, the number of texture sampling units is 32x2.

In graphic processors, another type of pipeline can be distinguished - vertex pipelines, they are responsible for calculating the geometric parameters of a three-dimensional image.

Now, let's look at the step-by-step, somewhat simplified process of pipeline calculation, followed by image formation:

1 - th stage.Texture vertex data goes to vertex pipelines, which calculate geometry parameters. At this stage, the “T&L” (Transform & Lightning) block is connected. This block is responsible for lighting and image transformation in three-dimensional scenes. Data processing in the vertex pipeline is carried out by the vertex shader program.

2 - oh stage.At the second stage of image formation, a special Z-buffer is connected to cut off invisible polygons and faces of three-dimensional objects. Next, the process of filtering textures occurs; for this, pixel shaders enter the “battle”. The OpenGL or Direct3D programming interfaces describe standards for working with three-dimensional images. The application calls a certain standard OpenGL or Direct3D function, and shaders perform this function.

3rd stage.At the final stage of image construction in pipeline processing, the data is transferred to a special frame buffer.

So, we have just briefly reviewed the structure and operating principles of GPUs; the information, of course, is not “easy” to understand, but for general computer development, I think it will be very useful :)

GPU (Graphics Processing Unit) is a processor designed exclusively for graphics processing and floating point calculations. It primarily exists to ease the workload of the main processor when it comes to demanding games or 3D graphics applications. When you play a game, the GPU is responsible for creating graphics, colors, and textures, while the CPU can handle artificial intelligence or game mechanic calculations.

What do we look at first when choosing a smartphone? If we ignore the cost for a moment, then first of all we, of course, choose the screen size. Then we are interested in the camera, the amount of RAM, the number of cores and the processor frequency. And here everything is simple: the more, the better, and the less, the worse. However, modern devices also use a graphics processor, also known as GPU. What it is, how it works and why it is important to know about it, we will tell you below.

The GPU architecture is not very different from the CPU architecture, but it is more optimized for efficient graphics processing. If you force the GPU to do any other calculations, it will show its worst side.

Video cards that are connected separately and run at high power exist only in laptops and desktop computers. If we are talking about -devices, then we are talking about integrated graphics and what we call SoC (System-on-a-Chip). For example, the processor has an integrated Adreno 430 graphics processor. The memory it uses for its operation is system memory, while graphics cards in desktop PCs are allocated memory available only to them. True, there are also hybrid chips.

While a CPU with multiple cores runs at high speeds, a GPU has many processor cores that run at low speeds and do little more than compute vertices and pixels. Vertex processing mainly revolves around the coordinate system. The GPU processes geometry tasks by creating three-dimensional space on the screen and allowing objects to move within it.

Pixel processing is a more complex process that requires a lot of processing power. At this point, the GPU applies various layers, applies effects, and does everything to create complex textures and realistic graphics. Once both processes are processed, the result is transferred to the screen of your smartphone or tablet. All this happens millions of times per second while you play a game.

Of course, this story about the operation of the GPU is very superficial, but it is enough to get a good general idea and be able to carry on a conversation with friends or an electronics seller, or understand why your device gets so hot during the game. Later we will definitely discuss the advantages of certain GPUs when working with specific games and tasks.

Based on materials from AndroidPit

Not many users know that video cards can do much more than just display an image on a monitor. Using CUDA, Stream and other similar technologies, you can significantly increase the performance of your computer by taking on other than your own calculations. The operating principle will be described below.

To display continuous frames on the screen in any modern game, the computer requires good performance. It is worth assuming that modern video cards match the performance of the latest versions of processors.

It is worth noting that when the video adapter is idle and does not perform image processing, its capabilities remain unclaimed. To avoid such downtime and to be able to assign some responsibilities to it, which will reduce the load on the processor, it is necessary to use special computer acceleration options. Below there will be detailed instructions on the principles of operation of this technology, which can increase PC performance.

How does a video card increase the speed of a computer?

Only special applications can take advantage of the capabilities of video cards. These programs can be combined with a video card and use one of 4 physical acceleration technologies.

CUDA. This development was created by Nvidia Corporation. This technology can be used for complex computational manipulations and for editing videos and pictures.

Stream. This mechanical acceleration technology is similar to the first, but developed by video adapter manufacturer AMD.
Both of these technologies are supported by all operating systems except Mac OS, and are used only with video cards from a suitable manufacturer. Software creators are forced to do additional work so that the video cards of both developers can increase the speed of their applications. Below are technologies that can work with boards from both manufacturers.

OpenCL. This technology was released by Apple in 2008 and is supported by all operating systems and any software. However, today there are no applications to speed up your computer using this technology. In addition, in terms of productivity increases, OpenCL is significantly behind the first two technologies.

DirectCompute. This technology was built into DirectX 11 by Microsoft. But it can only work on Windows 7 and Vista operating systems, and then with a small package of applications.

What kind of performance boost does the video card provide?

The increase directly depends on the graphics adapter and the performance of other elements of the computer. The performance increase is determined by the utilities and the operations performed. On a modern average PC, high-quality video conversion speeds can increase by up to 20 times. But editing photos with filters and special effects can speed up three hundred times.

What influences the high productivity of CUDA and similar technologies?

When performing complex tasks, the CPU on the motherboard initially divides the process into several smaller ones, and then processes them sequentially. The resulting intermediate result is located in the small but fast processor memory. When memory compartments become full, files are moved to cache memory, which is also located in the processor. But it takes quite a lot of time to exchange information between the processor and RAM, so the speed is not very high.

Video cards can sometimes perform such manipulations much faster. This may be influenced by several circumstances. One of them is parallel computing. If it is necessary to carry out several such manipulations, some of them can be carried out by the graphics module together with the processor.

For example, when working with videos or pictures, the utility needs to change a huge number of pixels, and at the same time using repetitive methods. Especially for this, the graphics adapter has hundreds of small processors, which are called streaming ones.

In addition, fast memory access is required. By analogy with central processes, graphics adapters also have their own intermediate memory and RAM. But in this case, they have many high-speed memory registers, which significantly increases the speed of calculations.

How many streaming CPUs do video cards have?

This is affected by the processor model. For example, the GeForse GTX 590 has two Fermi modules, each of which has 512 streaming CPUs. One of the most powerful video cards from AMD - Radeon HD 6990 - is also equipped with a pair of modules, each of which has 1536 processors. But with all this, the HD 6990 is significantly inferior to the GTX 590 in speed.

How to run CUDA or Stream?

You should not run anything, since technologies are an element of the hardware of video cards. After the graphics adapter driver installs an application that supports some technology, then the speed of the computer will automatically increase. To get full performance, you need to install the latest driver.
It is worth noting that users of AMD video cards need to download and install the AMD Media Codec Package.

Why don't all utilities work with these technologies?

Until OpenCL is widely adopted, software creators will need to tailor each application to work with Nvidia and AMD video cards. But not every manufacturer will incur additional costs.

Additionally, not all applications have the ability to provide a constant stream of lightweight computations that can occur in parallel. This can work great in conjunction with video and graphics editing programs. For mailers or text editors, these technologies will not help much.

Super PC

For example, the Chinese Tianhe-1A PC has 7168 Nvidia graphics modules, which support excellent performance. At the same time, 2.5 trillion calculations are performed per second. This computer consumes 4 megawatts of energy. This is how much electricity a town of five thousand people consumes.

Can a graphics adapter replace a central one?

Such a replacement is impossible. The design of these processors is completely different. The CPU is a universal computing unit that has the ability to process and send information to other PC elements. In turn, video cards are highly targeted devices, despite the fact that they perform a small number of operations, but at the same time at high speed.

What's in the future: universal chips

To increase CPU performance, Intel and AMD are constantly adding cores to their processors. In addition, they are constantly adding new technologies that can increase the efficiency of computing operations and the ability to process information in parallel.

Compared to central processing units, video cards already have a large number of simple cores that can perform complex calculations very quickly.

But it turns out that the initial differences in the operating principles of the video card and CPU are gradually erased. Therefore, the development of a universal chip is very logical. Today, computer users can use the full potential of a video card without expensive graphics chips.

Modern processors from leading developers can currently demonstrate the ability to connect a graphics adapter and a CPU and work as one universal computing unit.

In any of the chips, the CPU and video card cores are placed on a single chip. This provides the ability to quickly divide computational manipulation between cores. These technologies used are called Intel Quick Sync and AMD App. At this time, there are already separate applications that use similar technology.

In general, this is all you need to know about the differences between a central processor and a video card. As can be seen from what has been written, the graphics processor is capable of performing some central operations, especially for modern computers with powerful video cards.