Nvidia geforce gtx 650. Detailed characteristics. New technologies allow significantly increased productivity and durability of the video recording machine

• GPU: Geforce GTX 650 (GK107)
• interface: PCI Express x16
• Robotic GPU Frequency (ROPs): 1110 MHz (nominal - 1058-1100 MHz)
• The frequency of robotic memory (physical (effective)): 1350 (5400) MHz (nominal - 1200-1400 (4800-5600) MHz)
• Exchange bus width with memory: 128 bits
• The number of counting blocks in the GPU / the frequency of robotic blocks: 2/1110 MHz (nominal - 2/1058-1100 MHz)
• Number of operations (ALU) in block: 192
• Total number of operations (ALU): 384
• Number of texture blocks: 32 (BLF/TLF/ANIS)
• Number of rasterization blocks (ROP): 16
• expand: 150 × 100 × 30 mm (the remaining value is the maximum thickness of the video card)
• Color of textolite: blue
• Energy saving (peak in 3D / in 2D mode / in sleep mode): 78/37/22W
• Outside nests: 2 × DVI (Dual-Link / HDMI), 1 × VGA (d-Sub), 1 × HDMI 1.4a
• Support for rich processor work: SLI (Software)

The card has 1024 MB of GDDR5 SDRAM memory, placed in 8 microcircuits (4 each on the skin side of the PCB).

Map of support for supplementary life, moreover one 6-pin socket.

About the cooling system.

We carried out a follow-up of the temperature regime for an additional new version of the EVGA PrecisionX utility (author A. Nikolaichuk AKA Unwinder) and took away the results.

Package.

Gigabyte Geforce GTX 650 OC 1024 MB 128-bit GDDR5 PCI-E

Installation and drivers

Test bench configuration:

Processor based computers Intel Core i7-3960X (Socket 2011):
• Intel processor Core i7-3960X (o/c 4GHz);
• systemic asus motherboard Sabertooth X79 on Intel chipsets X79;
• motherboard MSI X79A-GD45 (8D) based on Intel X79 chipset;
• RAM 8 GB DDR3 SDRAM Corsair 1866 MHz;
• Seagate Barracuda 7200.14 3TB SATA2 hard drive;
• WD Caviar Blue WD10EZEX 1TB SATA2 hard drive;
• 2 living units Enermax Platimax 1200 W.
• operating system Windows 7 64-bit; DirectX11;
• monitor Dell UltraSharp U3011 (30 ");
• AMD drivers version Catalyst 12.11beta; Nvidia version 306.97

VSync enabled.

synthetic tests

Synthetic test packages we have created can be downloaded here:

• D3D RightMark Beta 4 (1050) with a description on the website.
• D3D RightMark Pixel Shading 2 and D3D RightMark Pixel Shading 3- testing of pixel shaders versions 2.0 and 3.0, upgraded.
• RightMark3D 2.0 h short description: For Vista without SP1, for Vista with SP1.

As synthetic tests of DirectX 11, we tested applications from SDK packages from Microsoft and AMD, as well as the Nvidia demo program. First, HDRToneMappingCS11.exe and NBodyGravityCS11.exe from the DirectX SDK (February 2010).

We took and supplements of both types of video chips: Nvidia and AMD. h ATI Radeon The SDK was taken butt (the stink is also in the DirectX SDK). The demonstrative program of Nvidia company, also known as Island11, was also victorious.

Synthetic tests were carried out on the next critical days:

• Geforce GTX 650 GTX 650)
• Geforce GTX 650 Ti with standard parameters (given GTX 650 Ti)
• Geforce GTX 560 with standard parameters (given GTX 560)
• Geforce GTX 550 Ti with standard parameters (given GTX 550 Ti)
• Radeon HD 7770 with standard parameters (given HD 7770)
• Radeon HD 7750 with standard parameters (given HD 7750)

To match the results of the new model of the Geforce GTX 650 video card, the decisions were taken from the onset causes. Geforce GTX 650 Ti is the closest model to the GTX 650, based on the more strained graphics processor GK106 of the Kepler architecture. Pairing with GTX 550 Ti and GTX 560 cicavo through different generations and price ranges of video cards - we know how much the productivity of the model of the new line has increased, in line with the predecessors of the novelty.

The video cards of the competing company AMD were chosen for our tests to the fact that the Radeon HD 7750 is one of the competitors of the new product, although it may be three times lower in price, in line with the recommended price of the announced video card Nvidia Geforce GTX 650. And the other model is the Radeon HD 7770 - may already three times more price. In other words, there is no new direct competitor for the GeForce GTX 650, they have to be taken seriously against the appointment of the AMD video card.

Direct3D 9: Pixel Shaders Test

Texturing tests and filling (fillrate) in the 3DMark Vantage package are lower, and the first group of pixel shaders, as expected, includes different versions of pixel programs with a similarly low complexity: 1.1, 1.4 and 2.0, which are used in old even simpler for today's videos.

The tests are even easier for today's GPUs, and their speed often interferes with texturing performance or fillrate. To that, the stench of the building is far from showing all the possibilities of modern videos, but all the same game add-ons, Yakikh to tsikh pіr dosit rich. Judging by our previous order, productivity the rest of the video cards in these tests, it is mostly surrounded by fillrate, although the speed of the texture modules is also high.

The new model of the Nvidia video card in the last couple of years performed badly - betting tests, outperformed the Geforce GTX 560, although the GTX 650 Ti performed decently. However, it was clear that even this model can be in the basis of GK106 and is the leader in these tests. In order to compete with AMD's video cards, the new Geforce GTX 650 showed better results on the equal Radeon HD 7770 (better parity of the best subtests), and not the weaker HD 7750.

Well, I realized that the Geforce GTX 550 Ti was left behind, although the novelty could not outperform the old model in one of the folding tests. Let's look at the results of more folding pixel programs of advanced versions:

The Cook-Torrance test is more intensive in terms of numbers, and the freshness in new deposits in terms of the number of ALUs and their frequencies, and also in terms of the flexibility of TMU. While this test is historically better suited for AMD's graphics solutions, all new Geforce boards based on the Kepler architecture seem to perform better in the new.

The new product in Nvidia shows a result approximately in the middle between HD 7770 and HD 7750. Moreover, in the lighting test, there is more exposure to ALU speed, while in the Water test, speed in which there is more exposure to texturing, we are about the same. The bad result of today's hero in the particular Geforce GTX 650, although the leader in both tests of the older model GTX 650 Ti, won strongly.

As far as more old Nvidia video cards are worth, the GTX 650 could outperform the GTX 560 and GTX 550 Ti in the Water test (the novelty is clearly a speedy one in terms of texture settings), and in the Cook-Torrance test it outperformed them, giving up the GTX 560 through less mathematical power, but again support theories.

Direct3D 9: Testing Pixel Shaders 2.0

Tests of DirectX 9 pixel shaders are more foldable than the previous ones, stink close to what we are contagious in multiplatform games, and are divided into two categories. Let's look at more simple shaders in version 2.0:

• Parallax Mapping- know the method of overlaying textures by most of today's games, reporting descriptions in the article "".
• Frozen glass- Foldable procedural texture of frozen folding with Kernel parameters.

There are two variants of shaders: with an orientation on mathematical calculations and with the choice of the value of the textures. Let's look at mathematically intensive options, more promising from the point of view of future additions:

Tse universal tests, productivity in such deposits and in the flexibility of ALU blocks, and in the flexibility of texturing, they also have an important chip balance and the efficiency of counting programs. Past our investigations show that in these specific tasks the AMD architecture looks three times better than the Nvidia graphics architecture.

Prices apply for up to two Radeon models, as selected for today's pairing. The "Frozen Glass" test has a lot of latency in terms of mathematical productivity, and the Radeon HD 7770 is once again better than the latest from Nvidia. In this test, it shows the result on equal three times the HD 7750 and GTX 550 Ti and the GTX 560 and GTX 650 Ti, which may be a great mathematical challenge.

In another "Parallax Mapping" test, the new Nvidia video card showed performance close to the Radeon HD 7770, and showed noticeably faster, lower than the HD 7750. That performance in the older GTX 650 Ti was clearly slow. Vtim, I wanted to catch up with the GTX 560, but I didn’t get far. Let's take a look at these tests in modification with the advantage of vibrations and textures of mathematical calculations:

For the first time, the boards based on Nvidia chips have clearly improved, and the Geforce GTX 650 is now closer to the top AMD competitor in both tests and three times faster than the GTX 560. Although the older Nvidia model is still far ahead. The current chips of the company and in their tasks obviously work effectively, and the new video card Geforce GTX 650 in the tests shows even a good result.

Ale, everything was old-fashioned, with an emphasis on texturing and fillrate. Let's take a look at the results of two tests of pixel shaders, and even version 3.0, the most advanced of our tests of pixel shaders for Direct3D 9. These are the most impressive from the point of view of today's games on the PC, among many multiplatform ones. Tests are being tested, which are strongly focused on ALUs, texture modules, shaders and collapsible programs, and include a large number of decals:

• Steep Parallax Mapping- Significantly more "important" variety of parallax mapping technique, also described in the article "Modern terminology of 3D-graphics".
• Fur- a procedural shader that visualizes smartly.

And the q-axis of the test is no longer exchanged for the productivity of texture vibrators and is most dependent on the efficiency of the folding shader code. In the best DX9 tests with the first version of the RightMark package, Nvidia's video cards were the strongest in the early years, but in the rest of the architectures, AMD's solutions sped up a lot and now the stench itself wins in different price ranges.

Yak and this time. Although the Geforce GTX 650 Ti is more expensive, and in equal terms it showed the best result, but the Radeon HD 7770 is not so good. We have already guessed today the Geforce GTX 650 model, which is much worse than its older sister, and it can be explained from the point of view of theoretical indications. A wondrous result of the novelty - it won against everyone, the HD 7750 in one of the tests. How many times has a new product been released closer to the Radeon HD 7750, and not to the HD 7770.

It's also for those reasons why it happened that the GTX 650 was forced to replace the GTX 550 Ti, moreover, in Fur and Steep Parallax Mapping. It was clear that the GTX 560 and even more GTX 650 Ti were left far ahead. The only possible explanation is the defect in memory bandwidth, which, according to other theoretical indicators, the GTX 650 is clearly better than the GTX 550 Ti.

Direct3D 10: testing PS 4.0 pixel shaders (texturing, looping)

In another version of RightMark3D, two familiar to the PS 3.0 test Direct3D 9 have been updated, as if DirectX 10 were rewritten, as well as two new tests. In the first couple, the possibility of including self-shadowing and shader supersampling was added, which additionally increases the intensity on the video chip.

These tests measure the productivity of pixel shaders with cycles with a large number of texture vibrations (in most modes up to ten hundreds of vibrators per pixel) and similarly to a small ALU load. In other words, they combine the flexibility of texture vibrations and the efficiency of coloring in pixel shaders.

The first test of pixel shaders will be Fur. At the lowest settings, 15 to 30 texture selections from the height map and two selections from the main texture are selected in the new one. The Effect detail mode - “High” increases the number of vibes up to 40-80, the inclusion of “shader” supersampling - up to 60-120 vibes, and the “High” mode in combination with SSAA is controlled by the maximum “vago” - from 160 to 320 vibes from the height map.

Checking several modes without supersampling turned on, they are remarkably simple, and the difference in results in the “Low” and “High” modes can be approximately the same.

For this test, the productivity is greater in the world, in terms of the quantity and efficiency of TMU blocks, and the efficiency of folding programs. And the variant without supersampling has an additional impact on productivity, more efficient fillrate and memory bandwidth. The results when detailing the “High” level go up to the second time and lower, lower when “Low”.

At the head of procedural visualization, with a large number of texture variations, for a couple of generations of graphic architectures, AMD has not only speeded up the price of Nvidia boards, but with the release of GCN and soared forward. And at the same time, pay Radeon yourself as the leaders of such divisions, what can we say about the high efficiency of their programs. As far as we can see on the diagram, the Radeon HD 7770 emerged as the leader, overtaking the Geforce GTX 650 Ti.

And the axis we reviewed today, the Geforce GTX 650 model, seriously compromised, showing the weakest result, sacrificing the youngest payment for the Radeon bet, and navit GTX 550 Ti! Imovirno, in which case wine is either relatively low throughput of memory building or not a lot of fillrate (productivity of ROP blocks). Let's marvel at the result of this test, but let's turn on the "shader" supersampling, which will make the robot work a lot more: it is possible, in such a situation, to change, and the memory bandwidth with the fill rate will be less:

Alas, this time the situation is similar to the previous one, although the GTX 650 could still outperform the old GTX 550 Ti, the results fell more strongly. It seems that when supersampling is enabled, which quadruples the theoretical effort, the results of AMD motherboards have improved by a factor of three, equal to those of Nvidia's video cards, and the Radeon HD 7770 is still in the lead.

Tested by us this year, Nvidia's board outperformed the GTX 550 Ti, giving up both competitors: Radeon HD 7770 and HD 7750. AMD's chips obviously have the advantage in terms of total cost, which is worthy for better pixel count. It is also significant and strong program of the GTX 650 model of its older sister - GTX 650 Ti. The budget GK107 is clearly not a good one for the job.

The upcoming DX10 test is improving the performance of folding pixel shaders with cycles with a large number of texture vibrators and is called Steep Parallax Mapping. With low adjustments of vicorist wines, 10 to 50 texture selections from the height map and three selections from the main textures. When the important mode is enabled, self-shadowing increases the number of samples by a factor of two, and supersampling increases the number by a factor of two. The most foldable test mode with supersampling and self-shadowing select from 80 to 400 texture values, so you have more times in pure mode with simple mode. we check the back of the head simple options without supersampling:

Another Direct3D 10 pixel-shader test is no longer practical, since parallax mapping is widely used in games, and important options, such as steep parallax mapping, have been featured in rich projects for a long time, for example, in games of the Crysis and Lost Planet series. In addition, in our test, the supersampling cream, you can turn on self-shadowing, which will increase the amount of bandwidth on the video chip by about two times - this mode is called "High".

The diagram is already similar to the one in front (without SSAA enabled), although the old Geforce GTX 550 Ti model spent more in speed than others. Nvidia's solution in this test still performs better than competing AMD ones, and the new Geforce GTX 650 board in the updated D3D10-version of the test without supersampling still looks like its young super-nic in the Radeon HD 7750, not seeming about the HD 7770, Yaka became the best-width, having overtaken the GTX 650 Ti. We wonder how to change the inclusion of supersampling, which sounds like a stronger drop in speed on Nvidia boards.

Everything is new about the same, like in "Fur". When supersampling and self-shadowing are enabled, the task will appear even more important, if the two options are enabled, the more advantage on the map may be at a higher rate, causing a serious drop in productivity. The difference between the visual indicators of protesting video cards has changed, the inclusion of supersampling is indicated, as in the front view - the Radeon HD 7770 video card has clearly expanded its visual indicators.

The AMD model still remains the leader, and the axis of the heroine of our review continues to act on both Radeon boards. GTX 650 can't compete with Radeon in this test, so it's graphics Nvidia solutions in our D3D10 tests, pixel shaders are less efficient. Well, even though in the previous generation of the GTX 550 Ti, the novelty took away the win, and that's good. Ale Geforce GTX 650 Ti with Radeon HD 7770 far ahead.

Direct3D 10: PS 4.0 Pixel Shader Tests (Cumulative)

A couple of tests of pixel shaders are coming to try to achieve the minimum number of texture variations to reduce the productivity of TMU blocks. They win a large number of arithmetic operations, and beat the stench of the most mathematical productivity of videos, the number of arithmetic instructions in the pixel shader.

The first math test is Mineral. This is a test of folding procedural texturing, in which there are only two selections from texture data and 65 instructions of the sin and cos type.

The results of our boundary mathematical tests show more or less differences in frequencies and number of counting blocks, but with a difference in the efficiency of their victories. AMD architecture of the rest of the years in such tests had a great advantage over competing Nvidia video cards, and the Californian company in the Kepler architecture increased the number of stream processors, and the peak mathematical productivity of the new Geforce models significantly increased.

Cho mi i bachimo for the results of the first mathematical test. The speed of the video cards on the diagrams approximately confirms the theory, although there are some vindications. Previously, the Geforce GTX 650 showed a result that may be identical to one of its direct competitors in the Radeon HD 7750, which is more than the theory. Naturally, while the HD 7770 came first, the Geforce GTX 650 Ti fell short. The new GTX 650 even overtook the old GTX 550 Ti, a little less.

Let's take a look at another shader test, which is called Fire. Vіn is more important for ALU, and the texture vibrator is only one in a new case, and the number of instructions for the type of sin and cos is increased twice, up to 130. We wonder what changed with the increase in vanity:

And the axis of another mathematical test, the outstanding result of the new product from Nvidia, turned out to be significantly lower, and now both boards have won AMD families Radeon HD 7000 We confirm our old visnovo about those that the effectiveness of calculating the Kepler architecture in our other mathematical test is clearly lower, lower for the front Fermi architecture. Therefore, the new model GTX 650 has become the most important video card in the world, although the performance of the GTX 550 Ti is small, but for us the program is more important to direct competitors like AMD.

Direct3D 10: geometry shader tests

The package RightMark3D 2.0 has two tests of the speed of geometric shaders, the first option is called "Galaxy", the technique is similar to "point sprites" earlier versions Direct3D. A new system of particles is animated on the GPU, a geometric shader from skin points creates chotiri vertices that fix the particle. Similar algorithms are responsible for the wide variation in future DirectX 10 games.

Changes in balancing in the tests of geometric shaders do not affect the end result of rendering, the picture is always absolutely the same, the way of processing the scene changes. The "GS load" parameter determines in which shaders the calculations are carried out - in the top or geometric one. Kіlkіst zavzhdi zavzhdi however.

Let's look at the first version of the "Galaxy" test, with calculations in the vertex shader, for three equal geometric folding:

Spіvіdnostnja shvidkostі with raznoї geometrії folded scenes is approximately the same for all solutions, productivity depends on the number of points, with skin crocodile FPS mayzhe dvorazova. The task for today's video cards is not too complicated, and the productivity in it is surrounded by the speed of processing the geometry and the throughput of the memory.

You can clearly see the difference between the results of video cards based on Nvidia chips and the AMD solution - in the presence of visibility in geometric pipelines. Although in previous tests with pixel shaders AMD boards were noticeably more efficient and smarter, the first geometry test showed that Nvidia boards are leading in such tasks.

I want today's new Geforce GTX 650 to have fewer geometric blocks, decently fit in the GTX 650 Ti, but don't care about the fight against its rightful competitors - the Radeon won't get around with a margin, so it's just like the GTX 550 Ti for now. We wonder how the situation changes when the parts are transferred to the geometric shader:

When changing the voltage in this test, the numbers did not change for the old Nvidia board and slightly increased for AMD boards and the current generation Nvidia solution. All video cards in this test react weakly to changes in the GS load parameter, which results in the transfer of part of the calculation to the geometric shader, and therefore all the visnovkas are too much. The new model Geforce GTX 650 is all so much better, lower GTX 650 Ti, but with the same swidshe both video cards Radeon families HD 7000 Cicavo, which changes in the offensive test, which conveys great interest in the geometric shader itself.

"Hyperlight" is another test of geometric shaders that demonstrates a variety of different techniques: instancing, stream output, buffer load. New victorious dynamically created geometry for additional implementation into two buffers, as well as new opportunity Direct3D 10 - stream output. The first shader generates a direct change, a speed and a direct change to its growth, the given data is sent to the buffer, which is picked up by another shader for creation. According to the skin points, there will be 14 peaks along the stake, up to a million exit points.

A new type of shader program is programmed to generate “changes”, and with the “GS load” parameter set to “Heavy” - another for their creation. Otherwise, it seems that in the “Balanced” mode, geometric shaders are only used for folding and “growing” changes, the visnovo is used for the additional “instancing”, and in the “Heavy” mode, the geometric shader is also involved in the visnovka.

It can be seen that the results of other solutions in different modes also approximately show changes in the vanity: in all modes, the productivity scales badly and is close to the theoretical parameters, for some skins the Polygon count attack is to blame for less or less than twice as much.

The speed of rendering in this test is also limited by the main geometric performance. And this time the Geforce GTX 650 showed speed closer to the model of the Fermi family and the competing Radeon, lower to the older model GTX 650 Ti. Vaughn more than a trio overtook them all, having sown another place. But the numbers can seriously change on the next diagram, in the test with more active multiple geometry shaders. Also, you will have to compare one by one the results, otrimani in the "Balanced" and "Heavy" modes.

Well, the results of the Radeon boards were close to the speed of the Geforce GTX 550 Ti, and the whole trio became the biggest in the ranking. For this test, the speed of processing geometry is important, for which Nvidia can do it clearly faster, and in Geforce budget chips, geometric blocks are not so rich, so the Geforce GTX 650 Ti based on GK106 clearly outperforms all other solutions presented in the diagram.

The speed in this test is clearly in line with the productivity of geometric blocks, and the result is similarly low new pay on the basis of the GK107 chip, it is clear that there is only one rasterization engine and two Polymorph engines, which roughly indicates that a competitor can pay. What can give a larger number of geometric blocks, is well seen from the result of the GTX 650 Ti, the chip can be at least double the number of blocks in rasterization and processing of geometry.

Direct3D 10: Vertex Shader Texture Selection Width

The "Vertex Texture Fetch" tests show a large number of texture variations from the vertex shader. The tests are essentially similar, so the correlation between the results of the cards in the "Earth" and "Waves" tests can be approximately the same. In both tests, displacement mapping is achieved on the basis of these texture vibrators, the only difference is that in the “Waves” test, smart transitions are distinguished, but in “Earth” it is not.

Let's take a look at the first "Earth" test, back in the "Effect detail Low" mode:

Our previous studies have shown that both texture speed and memory bandwidth can be added to the test results, especially in easy mode. And the results of some Nvidia video cards are still incomprehensible. And between boards similar in class, the difference in this test is usually not always small.

How many times did it sound unspeakable. Geforce GTX 550 Ti in light mode showed clearly amazing results, although everything is in order in folding. Geforce GTX 650 Ti became the best seen, and from all other motherboards (including the new GTX 650, as well as HD 7750 and HD 7770) they are close one to one, although the speed of the novelty can be blamed for indiscretions. We marvel at the productivity in this test from the greater number of texture vibrators:

On the other hand, the distribution of cards on the diagram has changed a little - the results of AMD's decisions have obviously gone down, and now the Radeon has received a new Geforce GTX 650, although it's not too bad. The speed of the GTX 550 Ti is squishy in light mode, but good in heavy mode. Well, our new product looks better than both AMD company boards and strongly looks like its older sister - GTX 650 Ti.

Let's take a look at the results of another test of texture selections from vertex shaders. The “Waves” test is tested by a smaller number of vibrations, then smart transitions are made into a new number. The number of white textural vibrations in this case is up to 14 (“Effect detail Low”) or up to 24 (“Effect detail High”) per skin vertex. Folding geometry changes similarly to the front test.

The results in another test of vertex texturing "Waves" are not similar to those in the front diagrams. For once, for some strange reason, the new model Geforce GTX 650 has shown close to the speed of the GTX 550 Ti and these boards can be called the most popular among the presentations. The speed of the Radeon HD 7750 is about the same as that of the GTX 650 Ti, while the older HD 7770 is more swedish (light mode). Let's look at another variant of this test:

With the complication of tasks, all decisions of Nvidia have become only better. Geforce video cards suffered especially strongly in light modes. Tse allowed the Radeon HD 7770 to become the clear leader of the division, the best solutions. The results of today's novelty were closer to the speed of the older GTX 650 Ti, but both are not very ryatu - the Radeon is still faster, especially in folding minds. Imovirno, this failure can be explained by shortcomings in the drivers, or due to the low memory bandwidth.

3DMark Vantage: Feature test

Synthetic tests from the 3DMark Vantage package will show us those that we missed earlier. Feature tests from the first test package can support DirectX 10 and keep in mind that they are updated in ours and are up to date. When analyzing the results new video cards Nvidia in this package, we were able to find some new and different colors, as they appeared before us in the tests with packages of the RightMark family. Feature Test 1: Texture Fill

The first test is a test of the firmness of the texture vibrators. The rectangle is filled with values ​​that are read from the small texture to the variations of the numerical texture coordinates, which change the skin frame.

The test of the Futuremark company, although it does not show the theoretically possible level of productivity of texture vibrators, but the efficiency of AMD and Nvidia video cards can still be high and the numbers of models should be close to the theoretical parameters. The new model Geforce GTX 650 looked like the GTX 650 Ti three times more than that, a bit small in theory. The oldest model of Nvidia and according to the theory of yaknaishvidsha in terms of texture and in which test it turned out that way.

Shcho Stoskhnnyannnya Shvidkosti GeForce GTX 650 s with a feck of competitors (Radeon HD 7750 I HD 7770) all the same Bilsh -Menhdovіdi - novelty nvidia in the textural skid of one of their own competitors. Geforce GTX 550 Ti from the previous generation, the new motherboard, although not significantly, has overtaken it. Feature Test 2: Color Fill

Check the speed of filling. It's just a simple pixel shader that doesn't cut performance. The interpolated color value is written to the screen-by-screen buffer (render target) with alpha-blending variables. There is a 16-bit trans-screened buffer in FP16 format, the most frequently hacked in games, like HDR rendering, so this test is quite common.

The ROP block performance test shows a surprising result. We have previously indicated that the numbers of the 3DMark Vantage subtest show the productivity of ROP blocks by improving the value of video memory bandwidth (so-called “Effective fillrate”), and the test often shows less memory bandwidth, but ROP productivity. But let's put it in the same way, the result of the Geforce GTX 650 while the GTX 650 Ti is not reasonable - theoretically, the difference is to blame but less.

As much as the compatibility between the GTX 650 and AMD's motherboards is supported by AMD, the model presented today showed a high level of speed with improved memory bandwidth, outperforming both competitors, so the Radeon HD 7750 and HD 7770 may have the same memory bandwidth. In this case, the novelty has no problems with the exchange of global productivity with the throughput of video memory, as it is used in games. And it seems that the GTX 650 in this case may have enough memory bandwidth. Feature Test 3: Parallax Occlusion Mapping

One of the most famous feature-tests, as the technique is already winning in games. At the new one, one chaotic pattern (more precisely, two tricout patterns) is made due to the special technique of Parallax Occlusion Mapping, which imitates the folding geometry. Vykoristovuyutsya to dosit resource operations on the transfer of exchanges and a map of the depths of high-rise buildings. Also, the surface is shaded for the help of the important Strauss algorithm. This test is more foldable and important for a video chip pixel shader, which is to compensate for numerical texture vibrations during tracing, dynamic de-scaling and folding shading of lighting according to Strauss.

The test is reviewed in the light of what we have done earlier, because the results in the new lie not only in terms of the mathematical calculations, the efficiency of the design, but also the flexibility of the textural vibrations, but richly in the light of what. To achieve high speed, the correct balance of the GPU is important here, as well as the efficiency of folding shaders. In this case, mathematical and textural productivity is more important, and in the 3DMark Vantage synthetics, the Geforce GTX 650 card was greatly outperformed by the GTX 650 Ti, moreover, by theory, it's about the same.

Although the model based on the GK106 chip became the best in the region, the Radeon HD 7770 turned out to be rather poor. Well, the Geforce GTX 650, although it wants to play against the mid-range video card of the forward GTX 550 Ti series, can not compete with the weakest Radeon from AMD. Unfortunately, the inexpensive Nvidia solution in such computational tasks is still better than the AMD solution. Feature Test 4: GPU Cloth

The test of cicavia tim, which develops physical interactions (imitation of tissue) for an additional video chip. Victory of the upper simulation, for the additional combined work of the vertex and geometric shaders, with decal passes. Use stream out to transfer vertices from one simulation pass to the next. In this way, we test the performance of vertex and geometry shaders and stream out speed.

The speed of rendering in this test also depends on the number of parameters, but the main factors influencing are the efficiency of processing geometry and the efficiency of geometric shaders, the situation on the diagram is different. Through the insertion of geometric blocks, you can see that all video cards manufactured by Nvidia should do it badly in the test, cheating for the price of paying Radeon. Apparently, only the Geforce GTX 550 Ti comes with both Radeon, but not much.

And the axis we reviewed today is the Geforce GTX 650 model, this time it easily outperforms both its competitors from AMD. In this vantage point low speed Radeon is quite clear, since clock frequency and the number of geometric blocks are important in geometric calculations. This is one of these tests, in which you can see the success of Nvidia's solution, which can improve the geometric pipeline. Before speech, the novelty based on the GK107 chip did not look much like the older sister GTX 650 Ti, based on the GK106. Feature Test 5: GPU Particles

Test of physical simulation of effects on the basis of systems of particles, which are covered for an additional video chip. The top simulation is also victorious, the skin top is a single part. Stream out victorious with the same method, which is in the front test. Hundreds of thousands of particles are being redeemed;

Similar to one of the tests of our RightMark3D 2.0, which are often rendered using an additional geometric shader, which creates some vertices from a skin point, which is often used. And the test is the most occupied shader blocks with vertex expansions, and stream out is also tested.

Just one more geometric test with 3DMark Vantage, and most of the drawings with these diagrams were completed a long time ago - the main test, in which the new chips based on the Kepler architecture are seriously sacrificed to the shortest representatives of the Fermi architecture. In today's fashion, the Geforce GTX 550 Ti appeared on the same level as the GTX 650 Ti and clearly outperformed it like a heroine of a low-level look, and її competitors.

In order to improve the speed of the new Geforce GTX 650 with the productivity of its supernics, then, regardless of the position of the AMD company, the new Nvidia still lost the pricey Radeon HD 7770, although the young model does not outperform them. In a nutshell, synthetic tests of imitation of fabrics and particles from the 3DMark Vantage test package, in which geometric shaders are actively used, show that Nvidia motherboards are stronger in them and generally outperform supernics in such tasks. Feature Test 6: Perlin Noise

The remaining feature-test of the Vantage package is a mathematical-intensive test of the video chip, which is based on the expansion of octaves for the Perlin noise algorithm in the pixel shader. Leather color channel victorious control noise function for greater focus on the video chip. Perlin noise - the same standard algorithm, often getting used to procedural textures, including rich mathematical patterns.

In a purely mathematical test from the package of the company Futuremark, which shows the peak productivity of video chips in boundary problems, we have even more subdivided the results, in parallel with two similar tests from our test package. In this way, the productivity solution is the same in the general agreement of the theory, and still differs from that, which we studied earlier in the mathematical tests from the RightMark 2.0 package.

Think AMD's Radeon, built on the basis of the GCN architecture chips, will do better with such tasks, and may always show better results in performance, if it's simple, but the math is even more intensive. Radeon HD 7770 shows excellent results, less than three times in a row shortest solution in por_vnyann_ - road models from the company Nvidia.

The video card Geforce GTX 650 reviewed this year showed the speed of both and more, lower Radeon HD 7750, but still not far off. For video cards based on the Kepler architecture chips, the efficiency in this respect is clearly higher, so the struggle with the price-related Radeon boards has become more successful, that model with the front line is clearly better at coping with the challenges.

Direct3D 11: Count Shaders

In order to protest the new solutions of Nvidia in tasks that win such new capabilities of DirectX 11, such as tessellation and enumeration shaders, we were quickened by butts from SDK packages and demo programs from Microsoft, Nvidia and AMD.

We can look at the back of the dough, like vikoristovuyut enumerate (Compute) shaders. Їх appearance - one of the most important innovations in the remaining versions of the DX API, the stench is already featured in today's games for other tasks: post-processing, simulation, etc. In the first test, the application of HDR rendering with tone mapping, DirectX SDK, with post-processing , which victorious pixel and enumerator shaders.

Although it is not the best example for calculating shaders, but it shows the difference in productivity in one of the specific tasks. There has been no difference in speed in the numbering and pixel shaders for all boards for a long time, as it was for video cards with GPUs of the forward architectures. Judging by our previous tests, the results in the task are clearly not only in terms of mathematical power and calculation efficiency, but also in terms of other factors, in terms of PSP and ROP productivity.

The novelty of Nvidia company in this test, it looks like the older model GTX 650 with the Ti index, but it shows a similar result compared to the speed of its competitors: Radeon HD 7770 and HD 7750. Another test of computational shaders is also taken from Microsoft DirectX SDK, it shows the problem of gravitation of N bodies (N-body) - a simulation of a dynamic system of particles, on which physical forces are added, such as gravity.

In this test, the results are absolutely logically superior to the Geforce GTX 650 Ti, and then the Radeon HD 7770. The young solutions of Nvidia and AMD go behind, and the new GTX 650 model shows perhaps the same result as the Radeon HD 7750 in this test. More similar to those in which the emphasis is placed on the PSP, and not on the speed of mathematical calculations. Let's move on to productivity tests in tessellation tasks, which traditionally can show the strongest side of the new board of the Geforce GTX 600 series.

Direct3D 11: Tessellation Performance

Numerical shaders are even more important, and another important innovation in Direct3D 11 is hardware tessellation. We looked at it in more detail in our theoretical article about the Nvidia GF100. Tessellation has already begun to be completed long ago in DX11 games, such as STALKER: Call to Come, DiRT 2, Aliens vs Predator, Metro 2033, Civilization V, Crysis 2, Battlefield 3 and others. Some of them use tessellation for character models, others - for imitation of realistic water surface or landscape.

Іsnuє kіlka rіznіh schemes rіzbіttya graphic primitives (tessellation). For example, phong tessellation, PN triangles, Catmull-Clark subdivision. So, the PN Triangles smashing scheme is victorious in STALKER: Call to Come, and in Metro 2033 - Phong tessellation. These methods are por_vnyano quickly and are simply used in the process of rozrobki іgor and іsnuyuchi engines, and have become popular.

The first test of tessellation will be the Detail Tessellation for the ATI Radeon SDK. The new one implemented not only tessellation, but two different techniques of pixel-by-pixel processing: simple overlay of normal maps and parallax occlusion mapping. Well, you can compare the DX11 solutions of AMD and Nvidia in different minds:

In a simple bumpmapping test, all the boards are unaware of what, and Nvidia boards show the result higher, lower AMD. The novelty of the Nvidia company does not give up to the older sister and trochs program to its super-nicks in the particular models Radeon HD 7770 and HD 7750, which demonstrates the unexpected closeness. And in the other subtests with more collapsible pop-pixel expansions, everything is bad for the GTX 650. These results show that the efficiency of collapsing mathematical calculations in pixel shaders is clearly higher for chips of the GCN architecture. As a result, the Radeon HD 7770 outperforms the GTX 650 Ti, and the HD 7750 outperforms the Geforce GTX 650 in a number of subtests.

In the most recent tessellation test on both Radeon video cards, the results are also strong. In this test, the tessellation of the patterning of tricots should be reduced and the speed in a new way does not impinge on the productivity of the geometry processing blocks. That's why the AMD company pays little in terms of productivity and their speed of processing jerseys as a whole, in order to show the results better than Nvidia. The Radeon HD 7770 and HD 7750 are again much closer and outperform both Nvidia's new product and older solution.

Another test of tessellation performance will be another example for 3D retailers with ATI Radeon SDK - PN Triangles. Vlasne, insults, but also enter the warehouse of the DX SDK, so we are convinced that on their basis they create their own code and game retailers. We protested against this butt with a different tessellation factor, in order to understand how strongly this change contributes to overall productivity.

At this buttstock, it is already more foldable geometry, and the geometrical power of different solutions for this test is to bring other visnovkas. Although all presented in the materials of the current solution can cope well with light and medium geometric challenges, showing high speed, but in important minds, Nvidia graphics processors are left with more productive ones, because AMD often corrected the step reduction methods (including rozbittya trikutnikov driver).

Announced today is the Geforce GTX 650 video card based on the GK107 chip, which is the smallest number of geometric blocks, in pairs with GK106 and even more GK104, it is worth remembering to do its older sister and to be known in the warehouses of competitors. Vtіm, obrazvі pay Radeon z vіdpovіdnogo pіnovogo dіpazon already look badly in all the minds of the tsy povnyannya.

Let's take a look at the results of one test - the demo program Nvidia Realistic Water Terrain, also known as Island. This demo features tessellation and displacement mapping for rendering realistically looking at the surface of the ocean and landscape.

Island is not a purely synthetic test for vimirovannaya all-key geometric performance of the GPU, so it is possible to vindicate and fold pixel and enumerate shaders, including, and so the aim is closer to real igor, in which all GPU blocks, and not just in the front ones, are victorious geometry tests.

We protested the demo program at different tessellation coefficients (in this case, the setting is called Dynamic Tessellation LOD). Якщо при найпершому коефіцієнті розбиття трикутників, коли швидкість не обмежена продуктивністю геометричних блоків, старша з відеокарт компанії AMD показує результат навіть краще обох Geforce, але при збільшенні геометричній роботи плати компанії Nvidia відразу ж починають вириватися вперед, а продуктивність Radeon помітно падає. Although AMD's streaming architecture has a geometric performance of polyp chips, the stench still continues to run on Nvidia's solutions in advanced synthetic tests.

Therefore, although the new budget board Geforce GTX 650 in all modes, the GTX 650 Ti based on the older GK106 chip is available, which is explained by a small number of active geometric blocks in the GK107 graphics processor, and a clear win over competing solutions in the Radeon 7775 HD 7775 HD 7775 HD manifestness, especially in the most sophisticated minds.

Visnovki according to synthetic tests

Результати синтетичних тестів нової моделі Geforce GTX 650, заснованої на графічному процесорі GK107 з сімейства Kepler, а також результати інших моделей відеокарт виробництва обох виробників дискретних відеочіпів показали, що нове рішення компанії Nvidia відповідає своєму ціновому діапазоні і за умови відповідних роздрібних цін є однією з найбільш in the distance video cards of the lower price segment.

For synthetic tests, the introduction of a new model in the form of Geforce GTX 650 Ti turned out to be great, but it is also explained by the need for different chips: GK107 and GK106 for GTX 650 and GTX 650 Ti, obviously. Well, and for a few tests, the new product performed badly in a row with the competing Radeon HD 7750 and HD 7770 models. games, better for everything, the new product will show the results closer to the younger Radeon, so the GK107 graphics processor is brought to budget solutions.

And inexpensive chips simply cannot but be short. The first of them ranks in a markedly smaller number of geometric blocks, in porous GK106, on which the GTX 650 Ti and GTX 660 models are based. important - there is simply not a great number of polygons there. In games, math and fillrate are more important, as well as memory bandwidth. І behind these indications GTX 650 is completely in its class - equal to Radeon HD 7750 and three times higher.

Zagal, based on theoretical data and synthetic tests, the Geforce GTX 650 video card is not a bad proposition for gravity and enthusiasts, if you want to pay three times more than $ 100 for a video card of a similar level. In the distance, the competitor's models Radeon HD 7750 and HD 7770 took off the superman, so they had to fight for the respect of buyers. In the next part of the article, we will check the productivity of the new solution in games - we will be amazed at the results of the Geforce GTX 650 in gaming add-ons.