• GPU: GeForce Titan (GK110)
• Interface: PCI Express x16
• GPU operating frequency (ROPs): 875-1020 MHz (nominal - 875-1020 MHz)
• Memory operating frequency (physical (effective)): 1750 (7000) MHz (nominal - 1750 (7000) MHz)
• Memory bus width: 384 bit
• Number of computational units in the GPU/block operating frequency: 15/875-1020 MHz (nominal - 15/875-1020 MHz)
• Number of operations (ALU) in block: 192
• Total number of operations (ALU): 2880
• Number of texturing units: 240 (BLF/TLF/ANIS)
• Number of rasterization units (ROP): 48
• Dimensions: 270×100×37 mm (the card occupies 2 slots in the system unit)
• PCB color: black
• Power Consumption (Peak 3D/2D/Sleep): 264/86/70 W
• Output Jacks: 1×DVI (Dual-Link/HDMI), 1×DVI (Single-Link/VGA), 1×HDMI 1.4a, 1×DisplayPort 1.2
• Multiprocessor support: SLI (Hardware)

The card has 3072 MB of GDDR5 SDRAM memory located in 12 chips on the front side of the PCB.

The card requires additional power in the form of two connectors: 8- and 6-pin.

About the cooling system.

Nvidia Geforce GTX 780 Ti 3072 MB 384-bit GDDR5 PCI-E
The cooling system is completely identical to the reference cooler from GTX Titan. The cooler has a traditional closed design with a cylindrical fan at the end. The radiator, pressed against the core, is based on an evaporation chamber, inside of which there is a special, easily evaporated liquid. The lower plate of the chamber is pressed against the core, heat is transferred to the liquid, which evaporates and carries the heat to the upper plate (which has cooling fins), where the vapors condense, etc. We have already talked more than once about this scheme for modern cooling of top-end accelerators. The fan drives air through the above-mentioned radiator and has a special impeller shape that gives a reduced noise level. We must say that at maximum load the noise is still slightly noticeable, because the maximum speed is above 2200 rpm. The memory chips are cooled by a central radiator (the cooler has a special plate that presses against the memory chips and power unit transistors).

We conducted a temperature study using the new version 4.2.1 of the EVGA PrecisionX utility (author A. Nikolaychuk AKA Unwinder) and obtained the following results.

After running the card for 6 hours under maximum gaming load, the maximum core temperature was 84 degrees, which is more than normal for such a powerful accelerator.

Equipment. The reference card arrived to us in OEM packaging, so there is no kit.

Installation and drivers

Test bench configuration:

• Computers based on Intel Core i7-3960X processor (Socket 2011):
  • 2 Intel Core i7-3960X processors (o/c 4 GHz);
  • WITH Hydro SeriesT H100i Extreme Performance CPU Cooler;
  • With Intel Thermal Solution RTS2011LC;
  • Asus Sabertooth X79 motherboard based on Intel X79 chipset;
  • MSI X79A-GD45(8D) motherboard based on Intel X79 chipset;
  • RAM 16 GB DDR3 Corsair Vengeance CMZ16GX3M4A1600C9 1600 MHz;
  • hard drive Seagate Barracuda 7200.14 3 TB SATA2;
  • hard drive WD Caviar Blue WD10EZEX 1 TB SATA2;
  • 2 SSD Corsair Neutron SSD CSSD-N120GB3-BK;
  • 2 Corsair CMPSU-1200AXEU power supplies (1200 W);
  • Corsair Obsidian 800D Full Tower case.
• operating system Windows 7 64-bit; DirectX 11;
• monitor Dell UltraSharp U3011 (30″);
• monitor Asus ProArt PA249Q (24″);
• AMD drivers version Catalyst 13.11beta8; Nvidia version 331.70 (for GTX 780 Ti) / 331/58 (for other Geforces)

VSync is disabled.

Synthetic tests

The synthetic test packages we use can be downloaded here:

• D3D RightMark Beta 4 (1050) with a description on the website 3d.rightmark.org.
• D3D RightMark Pixel Shading 2 and D3D RightMark Pixel Shading 3— tests of pixel shaders versions 2.0 and 3.0, link.
• RightMark3D 2.0 With brief description: under Vista without SP1, under Vista with SP1.

For synthetic DirectX 11 tests, we used examples from the Microsoft and AMD SDKs, as well as the Nvidia demo program. First, there are HDRToneMappingCS11.exe and NBodyGravityCS11.exe from the DirectX SDK (February 2010). We also took applications from both video chip manufacturers: Nvidia and AMD. The examples DetailTessellation11 and PNTriangles11 were taken from the ATI Radeon SDK (they are also in the DirectX SDK). Additionally, Nvidia's demo program, Realistic Water Terrain, also known as Island11, was used.

Synthetic tests were carried out on the following video cards:

• GeForce GTX 780 Ti GTX 780 Ti)
• GeForce GTX Titan with standard parameters (further GTX Titan)
• GeForce GTX 780 with standard parameters (further GTX 780)
• Radeon R9 290X with standard parameters in the “Uber Mode” mode (hereinafter R9 290X)
• Radeon HD 7990 with standard parameters (further HD 7990)

To analyze the results of the new high-end video card Geforce GTX 780 Ti, these solutions were chosen for the following reasons. Geforce GTX Titan is an exclusive model based on the same GK110 chip, has a large amount of video memory and is sold at a much higher price. Titan is Nvidia's previously powerful single-chip solution, and it will be interesting to see how much faster the new product turns out to be. A comparison with the GeForce GTX 780 will be interesting because this is a less expensive video card from the company, based on the same chip, but with a quarter fewer active execution units.

For our comparison, two video cards were chosen from competing company AMD, based on different graphics processors and even different numbers of them. At the time of the release of Nvidia's new product, the Radeon R9 290X is its closest competitor in price, and at the same time the most productive video card from AMD. And the Radeon HD 7990 has two Tahiti video chips at once and is not a competitor to the GTX 780 Ti, but we will be interested to see how the speed of such a powerful dual-chip solution compares with the best single-chip solution from Nvidia.

Direct3D 9: Pixel Shaders tests

We will look at texturing and fill rate tests from the 3DMark Vantage package a little later, and the first group of pixel shaders that we use includes various versions of pixel programs of relatively low complexity: 1.1, 1.4 and 2.0, found only in old games, very simple for modern video chips.

Modern GPUs cope with the simplest tests with ease; the speed of powerful solutions in them always rests on various limits, which is especially true for GeForce. These tests are not able to show the capabilities of modern video chips and are interesting only from the point of view of outdated gaming applications. The performance of modern video cards is often limited by the speed of texturing or fillrate, and Nvidia video cards have long ceased to be optimized for such tasks, as the results of today's comparison clearly show.

Look, all Geforce boards differ slightly in speed from each other, the difference between the GTX 780 Ti and Titan is only 1-4%, with a much higher theoretical one. The new video card model released today in this comparison, although it turns out to be the best among Nvidia cards, is clearly inferior to its main competitor, the Radeon R9 290X, which is always noticeably ahead. Let's look at the results of more complex intermediate pixel programs:

The Cook-Torrance test is more computationally intensive, and its speed depends more on the number of ALUs and their frequency, but also on the speed of the TMU. This test is historically better suited for AMD graphics solutions, although the new top-end GeForce boards based on the Kepler architecture also show strong results, which we can see from the generally good numbers of the new GeForce GTX 780 Ti.

The most powerful board from the GeForce GTX 700 family turned out to be 5-6% faster than the exclusive GTX Titan, which is also less than the theoretical difference and can only be explained by the emphasis on the performance of ROP units. Nvidia's new product slightly outperforms its main competitor in one of the tests - in the Water test, where texturing speed is more important, I'm not talking about mathematical performance, in which AMD boards have some advantage. Therefore, in the second test, the results of the GeForce GTX 780 Ti are slightly lower than those of the Radeon R9 290X. On average, there is clear parity in these tests.

Direct3D 9: pixel shader tests Pixel Shaders 2.0

These DirectX 9 pixel shader tests are more complex than the previous ones, they are close to what we now see in multi-platform games, and are divided into two categories. Let's start with the simpler version 2.0 shaders:

• Parallax Mapping- a method of texture mapping familiar to most modern games, described in detail in the article “”.
• Frozen Glass- a complex procedural texture of frozen glass with controllable parameters.

There are two variants of these shaders: those with a focus on mathematical calculations and those with a preference for sampling values ​​from textures. Let's consider mathematically intensive options that are more promising from the point of view of future applications:

These are universal tests, in which performance depends on both the speed of ALU units and the texturing speed; the overall balance of the chip and the efficiency of execution of computer programs are also important in them. Our past studies show that in these specific tasks, the GCN architecture from AMD is significantly better than the Nvidia Kepler graphics architecture, and this happened this time too.

In the Frozen Glass test, speed is more dependent on mathematical performance, and in the case of all Geforce boards there is always some kind of obstacle, due to which Nvidia boards lose almost twice as much to the almost better single-chip Radeon. The GeForce GTX 780 Ti model is only 1% faster than the GTX Titan, which only confirms the strange performance emphasis for all GeForces.

But in the second “Parallax Mapping” test, the new Geforce GTX 780 Ti video card showed performance 15% higher than that of the GTX Titan, which is already very close to theory. As for comparison with its competitor, the comparison of the new product with the rival model Radeon HD R9 290X is not the most rosy - the AMD board is faster in this test by almost a third. Let's consider these same tests in a modification with a preference for samples from textures over mathematical calculations:

In these conditions, the position of video cards produced by Nvidia has improved somewhat, because they traditionally cope with texture samples better than with mathematical calculations. But the Radeon R9 290X is still ahead of today's new product by a good margin, especially in the Frozen Glass test, where the difference remains indecent. The new product is 4-12% faster than the GTX Titan, which is more or less in line with theory. As for comparison with the R9 290X, the GTX 780 Ti is only close to it in the Parallax Mapping test, and even then the difference exceeds 20%.

However, these were long-outdated tasks, with an emphasis on texturing, which is almost never seen in games. Next we will look at the results of two more pixel shader tests, but this time version 3.0, the most complex of our pixel shader tests for Direct3D 9. They are more indicative from the point of view of modern games on PC, including many multi-platform ones. The tests differ in that they heavily load both the ALU and texture modules; both shader programs are complex and lengthy and include a large number of branches:

• Steep Parallax Mapping- a much more “heavy” type of parallax mapping technique, also described in the article “Modern terminology of 3D graphics”.
• Fur— a procedural shader that renders fur.

These tests are no longer limited by the performance of texture samples or fill rates alone, and the speed in them most of all depends on the efficiency of the execution of complex shader code. In the most difficult DX9 tests from the first version of the RightMark package, Nvidia video cards were slightly stronger in previous years, but the GCN architecture helped AMD video cards take the lead at least in the complex parallax mapping test, especially after carefully fine-tuning the Catalyst drivers.

Nvidia's top new product shows very good results in these tasks, outperforming the best of its predecessors based on the same GK110 chip by 11%, which is close to the theoretical figures for the difference in mathematical performance. As for comparison with the most powerful top-end graphics card based on the Hawaii chip from a competitor, the GTX 780 Ti lags behind it only in the parallax mapping test. But in the Fur test, the new Radeon R9 290X still lost to the Geforce GTX 780 Ti, although not by that much. In general, the situation in these tests is ambiguous.

Direct3D 10: PS 4.0 pixel shader tests (texturing, loops)

The second version of RightMark3D included two already familiar PS 3.0 tests for Direct3D 9, which were rewritten for DirectX 10, as well as two more new tests. The first pair added the ability to enable self-shadowing and shader supersampling, which further increases the load on video chips.

These tests measure the performance of pixel shaders running in cycles with a large number of texture samples (in the heaviest mode, up to several hundred samples per pixel) and a relatively small ALU load. In other words, they measure the speed of texture samples and the efficiency of branches in the pixel shader.

The first test of pixel shaders will be Fur. At the lowest settings, it uses 15 to 30 texture samples from the height map and two samples from the main texture. The Effect detail mode - “High” increases the number of samples to 40-80, the inclusion of “shader” supersampling - up to 60-120 samples, and the “High” mode together with SSAA is characterized by maximum “heaviness” - from 160 to 320 samples from the height map.

Let's first check the modes without supersampling enabled; they are relatively simple, and the ratio of results in the “Low” and “High” modes should be approximately the same.

Performance in this test depends on the number and efficiency of TMUs, as well as the efficiency of executing complex programs. And in the version without supersampling, the effective fill rate and memory bandwidth also have an additional impact on performance. The results at the “High” level of detail are up to one and a half times lower than at the “Low” level.

In tasks of procedural fur visualization with a large number of texture samples, over a couple of generations of graphic architectures, AMD has reduced the difference with Nvidia boards, and with the release of video chips based on the GCN architecture, it has completely taken the lead, and now Radeon boards are the leaders in these comparisons, which indicates high efficiency of their implementation of these programs.

The new top-end GeForce GTX 780 Ti is 11-12% ahead of the exclusive GTX Titan model, beating other Nvidia solutions, which is in line with the theory. But, taking into account the fact that in this test even AMD boards of the previous generation are faster than the new GeForce GTX 780 series, there is no point in considering a comparison of the R9 290X and GTX 780 Ti - the AMD model shows too high a result, not to mention the dual-chip card of the previous generation, which became the fastest here.

Let's look at the result of the same test, but with shader supersampling enabled, which increases the work by four times: perhaps in this situation something will change, and memory bandwidth with fill rate will have less effect:

The situation is similar to what we saw in the previous diagram, but Nvidia video cards are even slightly inferior to their AMD rivals. The new GeForce GTX 780 Ti turns out to be faster than the GTX Titan model by up to 11%, which is close to the theoretical difference in mathematical performance. Unfortunately, the loss to its direct competitor in the form of the Radeon R9 290X is very impressive. It is again confirmed that AMD chips, which prefer per-pixel calculations, clearly have an advantage in such calculations.

The next DX10 test measures the performance of complex pixel shaders with loops with a large number of texture samples and is called Steep Parallax Mapping. At low settings it uses 10 to 50 texture samples from the height map and three samples from the main textures. Enabling heavy mode with self-shadowing doubles the number of samples, and supersampling quadruples this number. The most complex test mode with supersampling and self-shadowing selects from 80 to 400 texture values, that is, eight times more than the simple mode. Let's first check simple options without supersampling:

The second Direct3D 10 pixel shader test is more interesting from a practical point of view, since types of parallax mapping are widely used in games, and heavy options, like steep parallax mapping, have long been used in many projects, for example, in the games of the Crysis and Lost Planet series. In addition, in our test, in addition to supersampling, you can enable self-shadowing, which approximately doubles the load on the video chip - this mode is called “High”.

The diagram is generally similar to the previous one, also without SSAA enabled, and this time the GeForce GTX 780 Ti is ahead of the GTX Titan by as much as 16-18%, which is even more than the theoretical difference in ALU speed. Most likely, the speed here also depends on the memory bandwidth of the video memory. But since Nvidia video cards in this test always perform worse than competing solutions from AMD, the GeForce GTX 780 Ti model in the updated D3D10 version of the test without supersampling again shows a worse result than the Radeon R9 290X, not to mention the dual-chip HD 7990. Let's see what difference enabling supersampling will make:

Everything is again approximately the same as in “Fur” - when supersampling and self-shadowing are enabled, the task becomes even more difficult; enabling two options together increases the load on the cards by almost eight times, causing a serious drop in performance. The difference between the speed performance of the tested video cards has changed only slightly; turning on supersampling has less of an impact than in the previous case.

We again see that Radeon graphics solutions perform more efficiently in our D3D10 pixel shader tests compared to competing GeForces, and the older top-end board on the Hawaii chip outperforms the GeForce GTX 780 Ti announced today by a huge advantage. Compared to other Nvidia motherboards, the new product shows better performance, outperforming the GTX Titan by 10-11%, which is approximately what it should be according to theory. It is clear that the GTX 780 is even further behind. Let's see what happens in purely computational problems.

Direct3D 10: PS 4.0 Pixel Shader Tests (Compute)

The next couple of pixel shader tests contain a minimum number of texture fetches to reduce the performance impact of the TMU units. They use a large number of arithmetic operations, and they measure precisely the mathematical performance of video chips, the speed of execution of arithmetic instructions in a pixel shader.

The first math test is Mineral. This is a complex procedural texturing test that uses only two samples of texture data and 65 sin and cos instructions.

The results of limiting mathematical tests usually only approximately correspond to the difference in frequencies and the number of computational units; they are influenced by the different efficiency of their use in specific solutions, and driver optimization is also important. In the case of the Mineral test, the new GeForce GTX 780 Ti model is only 8% ahead of the GTX Titan, which is clearly lower than the theoretical difference in mathematical performance between them. Probably some kind of limitation is affecting it, because this cannot be explained by the difference in characteristics.

As we already know, AMD architectures have always had a significant advantage over competing Nvidia solutions in such tests, but with the Kepler architecture, the Californian company managed to increase the number of stream processors, and the peak mathematical performance of GeForce models, starting with the GTX 680, has increased significantly. We can see this from the results of our first mathematical test, where the best GeForce video card, although still inferior to the board based on the Hawaii chip, is only 9% ahead of its competitor GTX 780 Ti. However, judging by the prices, the Nvidia graphics card should be ahead, so there is still some work to be done.

Let's look at the second shader calculation test, which is called Fire. It is heavier for an ALU, and there is only one texture fetch, and the number of sin and cos instructions has been doubled, to 130. Let's see what has changed with increasing load:

But in the second mathematical test we see completely different results from video cards relative to each other. The difference between the GTX Titan and today's new product in this test was even a little more theoretical - 19%. This looks much more like a true difference in math performance.

Unfortunately, even with such a strong result, the new single-chip top Nvidia Geforce GTX 700 series cannot cope with its competitor from AMD, which also has a lower price. The GeForce GTX 780 Ti cannot compete with the latest AMD board, which turns out to be 12% faster in the second mathematical test. The only good thing is that the GTX 780 Ti is clearly faster than the GTX 780 and Titan.

Direct3D 10: geometry shader tests

The RightMark3D 2.0 package has two geometry shader speed tests, the first option is called “Galaxy”, a technique similar to “point sprites” from previous versions of Direct3D. It animates a particle system on the GPU, a geometry shader from each point creates four vertices that form a particle. Similar algorithms should be widely used in future DirectX 10 games.

Changing the balancing in geometry shader tests does not affect the final rendering result, the final image is always exactly the same, only the methods of processing the scene change. The “GS load” parameter determines which shader the calculations are performed in - vertex or geometry. The number of calculations is always the same.

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

The ratio of speeds for different geometric complexity of scenes is approximately the same for all solutions, performance corresponds to the number of points, with each step the FPS drop is close to twofold. This task is not very difficult for modern video cards, and performance is limited by the speed of geometry processing, and sometimes by memory bandwidth.

There is some difference between the results of video cards based on Nvidia and AMD chips, due to differences in the geometric pipelines of the chips from these companies. If in previous tests with pixel shaders AMD boards were noticeably more efficient and faster, then geometry tests show that Nvidia boards are more productive in such tasks, even despite the increase in the number of geometry blocks in Hawaii.

But the difference between AMD and Nvidia is no longer as great as it used to be. Nvidia's geometric performance solutions have always done better and are therefore faster. Today's new GeForce GTX 780 Ti turns out to be approximately equal in performance to the earlier solution in the form of the GTX Titan, which indicates testing the performance of the geometric pipeline. Let's see how the situation changes when we transfer part of the calculations to the geometry shader:

As the load changed in this test, the numbers improved slightly for both AMD and Nvidia boards. The video cards in this test of geometry shaders react weakly to changes in the GS load parameter, which is responsible for transferring part of the calculations to the geometry shader, so all the conclusions remain the same. The new Geforce GTX 780 Ti model still shows performance on par with other boards based on the GK110 chip. And the competing Radeon R9 290X still lags behind them, so nothing changes in the conclusions.

“Hyperlight” is the second test of geometry shaders, demonstrating the use of several techniques at once: instancing, stream output, buffer load. It uses dynamic geometry creation using dual-buffer rendering, as well as a new feature of Direct3D 10 - stream output. The first shader generates the direction of the rays, the speed and direction of their growth, this data is placed in a buffer, which is used by the second shader for drawing. For each point of the ray, 14 vertices are built in a circle, up to a million output points in total.

A new type of shader programs is used to generate “rays”, and with the “GS load” parameter set to “Heavy”, also to draw them. In other words, in the “Balanced” mode, geometry shaders are used only to create and “grow” rays, the output is carried out using “instancing”, and in the “Heavy” mode, the geometry shader is also involved in output.

Unfortunately, “Hyperlight” simply does not work on all modern AMD video cards, including the top-end Radeon R9 290X. At some point, another driver update resulted in this test simply not running on boards from this company. And therefore, the most interesting geometry test of our package, which assumes a heavy load on geometry shaders, cannot say anything about comparing AMD and Nvidia boards.

But we can at least see what has changed in the case of Nvidia solutions. The relative results of solutions in different modes approximately correspond to the change in load: in all cases, performance scales well and is close to theoretical parameters, according to which each subsequent level of “Polygon count” should be slightly less than twice as slow.

The rendering speed in this test is limited mainly by geometry performance, but in the case of balanced loading of geometry shaders, all results are close. The Geforce GTX 780 Ti showed a speed 6-8% higher than the Titan level, which suggests that the matter is clearly not only in geometric performance. However, the numbers may change significantly in the next diagram, in a test with more active use of geometry shaders. It will also be interesting to compare the results obtained in the “Balanced” and “Heavy” modes with each other.

The most important parameter in this test is geometry processing speed, which Nvidia does very well, especially with the fully unlocked GK110 chip on which the Geforce GTX 780 Ti model in question is based. Due to the larger number of geometric blocks, the GeForce GTX 780 Ti outperforms the GTX Titan by 14-19%, and the latter, in turn, is significantly faster than the younger board based on the GK110 chip - the GTX 780.

Direct3D 10: texture fetching speed from vertex shaders

The Vertex Texture Fetch tests measure the speed of a large number of texture fetches from the vertex shader. The tests are essentially similar, so the ratio between the cards' results in the Earth and Waves tests should be approximately the same. Both tests use displacement mapping based on texture sample data, the only significant difference is that the “Waves” test uses conditional branches, while the “Earth” test does not.

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

Previous research has shown that the results of this test can be affected by both fill rate and memory bandwidth, which is especially noticeable in easy mode. The results of Nvidia graphics cards are often limited by something strange, as evidenced by the similar results of all graphics cards based on the GK110 GPU.

As expected, the fastest among single-chip solutions in comparison was the top-end Radeon R9 290X, and the new GeForce GTX 780 Ti presented today is inferior to it in all modes, even in heavy mode, where the difference is least. Nvidia's new top-end board outperformed the GTX Titan in this test by 10-13%, which is close to theory. Let's look at the performance in the same test with an increased number of texture samples:

The situation on the diagram has changed significantly - the results of AMD solutions in heavy modes have worsened, while for GeForce they have remained in almost the same positions. Now the Radeon R9 290X shows results noticeably higher than the speed of the new Nvidia product only in the simplest mode, and in medium and heavy mode the Geforce GTX 780 Ti announced today is ahead of it. The difference between the GTX 780 Ti and the GTX Titan is 9-12%, which is in line with theory.

Let's look at the results of the second test of texture fetches from vertex shaders. The Waves test has a smaller number of samples, but it uses conditional jumps. The number of bilinear texture samples in this case is up to 14 (“Effect detail Low”) or up to 24 (“Effect detail High”) per vertex. The complexity of the geometry changes similarly to the previous test.

The results in the second "Waves" vertex texturing test are generally similar to what we saw in the previous charts. For some reason, the performance of all GeForce boards based on GK110 in light mode remains greatly underestimated, and they are almost twice as bad as the speed of the dual-chip Radeon HD 7990. The speed of the new top-end GeForce GTX 780 Ti compared to its counterparts in this test is not bad, the new the single-chip top based on GK110 turned out to be 8-10% faster than the GTX Titan. Let's consider the second version of the same test:

In the second test of texture samples with increasing complexity of the task, the speed of all solutions became lower, and GeForce video cards were especially seriously affected in light modes. The results of today's new GeForce GTX 780 Ti from Nvidia were only 5% better than the GTX Titan based on the same chip, which suggests that the main performance limit in this test for Nvidia video cards is the performance of the ROP units, most likely .

3DMark Vantage: Feature tests

Synthetic tests from the 3DMark Vantage package will show us what we previously missed. Feature tests from this test package support DirectX 10 and are interesting in that they differ from ours and are still relevant. Probably, when analyzing the results of the new Geforce GTX 780 Ti video card in this package, we will draw some new useful conclusions that eluded us in tests from the RightMark family of packages.

The first test measures the performance of texture fetch blocks. This involves filling a rectangle with values ​​read from a small texture using multiple texture coordinates that change every frame.

The performance of AMD and Nvidia video cards in the Futuremark texture test is quite high and the comparative figures of the models are close to the corresponding theoretical parameters. The older top model Geforce GTX 780 Ti, which was released today, in this test is only 2% faster than the formerly most powerful GTX Titan video card, which is not too close to theory, I must admit.

Naturally, the GTX 780 lags even further behind the two most expensive Nvidia solutions in terms of texturing speed. As for comparing the GeForce GTX 780 Ti with the solution of its competitor Radeon R9 290X, the new Nvidia product is slightly faster in texture speed than the board based on the Hawaii GPU. What was expected based on theoretical indicators.

The second task is a fill rate test. It uses a very simple pixel shader that does not limit performance. The interpolated color value is written to an off-screen buffer (render target) using alpha blending. The 16-bit off-screen buffer of the FP16 format is used, which is most often used in games that use HDR rendering, so this test is quite timely.

In this case, it is not the peak speed of ROP blocks that is measured, the numbers from the 3DMark Vantage subtest show the performance of ROP blocks taking into account the amount of video memory bandwidth (the so-called “effective fill rate”), and the test measures exactly throughput, not ROP performance.

Therefore, the result of the announced Nvidia board in the performance test of ROP units turned out to be 10% better compared to the GTX Titan, since there is a theoretical difference in memory bandwidth between them. The same applies to outperforming the competitor in the form of the Radeon R9 290X - in fact, the speed of the ROP blocks is higher on the AMD board, but due to lower memory bandwidth it loses to the new Geforce GTX 780 Ti.

One of the most interesting feature tests, since a similar technique is already used in games. It draws one quadrilateral (more precisely, two triangles) using a special Parallax Occlusion Mapping technique that simulates complex geometry. Quite resource-intensive ray tracing operations and a high-resolution depth map are used. This surface is also shaded using the heavy Strauss algorithm. This is a test of a very complex pixel shader that is heavy for a video chip, containing numerous texture samples during ray tracing, dynamic branching and complex lighting calculations according to Strauss.

This test of the 3DMark Vantage package differs from the ones we conducted earlier in that the results depend not solely on the speed of mathematical calculations, the efficiency of branch execution or the speed of texture samples, but on several parameters simultaneously. To achieve high speed in this task, the correct balance of the GPU is important, as well as the efficiency of executing complex shaders.

In this case, both mathematical and texture performance are important, and possibly also ROP speed, since in this “synthetics” from 3DMark Vantage, the new Geforce GTX 780 Ti is only 5% ahead of the more expensive Nvidia board, which does not quite correspond to the theoretical difference in texturing speed and computing performance.

If we compare the new product with its competitor's solution, then in this test the GTX 780 Ti cannot compete with the Radeon R9 290X, not to mention the dual-chip HD 7990, since AMD GPUs are more efficient in this particular task. Alas, the GTX 780 lags behind its closest competitor in price by 20%, which is quite a lot.

The fourth test is interesting because it calculates physical interactions (fabric imitation) using a video chip. Vertex simulation is used, using the combined work of vertex and geometry shaders, with several passes. Use stream out to transfer vertices from one simulation pass to another. Thus, the execution performance of vertex and geometry shaders and the stream out speed are tested.

The rendering speed in this test should also depend on several parameters at once, and the main influencing factors should be geometry processing performance and the efficiency of geometry shaders. But the picture on the diagram turned out to be very strange, both Radeon video cards show a frame rate of about 130 FPS, and the results of the three GeForces also hit the limit, but at a level of about 95-100 FPS, as we saw earlier.

And yet, the new product is 7% ahead of the expensive GTX Titan, oddly enough. The new top-of-the-range model from Nvidia shows speeds one-third worse than the competitor’s older board, the Radeon R9 290X. And all this despite the fact that the geometric performance of Nvidia video cards should be higher than that of competitor solutions, since they have a larger number of corresponding execution units. We will also recheck geometric performance in DirectX 11 tests.

Test of physical simulation of effects based on particle systems calculated using a video chip. Vertex simulation is also used, each vertex representing a single particle. Stream out is used for the same purpose as in the previous test. Several hundred thousand particles are calculated, all are animated separately, and their collisions with the height map are also calculated.

Similar to one of our RightMark3D 2.0 tests, particles are rendered using a geometry shader that creates four vertices from each point to form a particle. But the test mostly loads shader units with vertex calculations; stream out is also tested.

In the second geometry test from 3DMark Vantage, the situation has changed, and this time the clear leader is the dual-chip Radeon HD 7990, which is out of the standings today. Nvidia's new product was only 1% superior to the GTX Titan board based on the same GK110 chip, which indicates an emphasis on geometric performance, at least for Nvidia boards.

If we compare the speed of the new GeForce with its only competitor from AMD, the new board is very close to its rival - they both show similar results in this task. And this is a good result most likely for Radeon, because it costs less, and even before, synthetic tests for simulating fabrics and particles from the 3DMark Vantage test package, which actively use geometry shaders, showed that Nvidia boards are significantly ahead of competing models from AMD, but now everything is not so obvious.

The last feature test of the Vantage package is a mathematically intensive test of the video chip; it calculates several octaves of the Perlin noise algorithm in the pixel shader. Each color channel uses its own noise function to put more stress on the video chip. Perlin noise is a standard algorithm often used in procedural texturing and uses a lot of math.

In a purely mathematical test from the Futuremark package, showing the peak performance of video chips in extreme tasks, we see a different distribution of results compared to similar tests from our test package. In this case, the performance of the solutions does not quite correspond to the theory and is at odds with what we saw earlier in mathematical tests from the RightMark 2.0 package.

AMD's Radeon video cards, based on GCN architecture chips, cope very well with such tasks and show better results in cases where intensive "math" is performed. This does not apply except to the dual-chip Radeon HD 7990 board, which clearly did not work efficiently in this case. However, if we compare the GeForce GTX 780 Ti announced today with the Radeon R9 290X, the latter outperforms the Nvidia board by 18%.

The GTX 780 Ti video card released on the market today showed speeds even slightly slower than the GTX Titan model from the same manufacturer and based on the same chip, which absolutely does not correspond to the theory. Today's new product still outperformed the GTX 780 by 11%, although it should have won by a much greater margin. Probably, some limitation of GPU Boost had an effect, reducing the frequency of the GK110 in the GTX 780 Ti during the last synthetic test of the package.

Direct3D 11: Compute Shaders

To test Nvidia's new solution on tasks that use DirectX 11 features such as tessellation and compute shaders, we used samples from the SDKs and demos from Microsoft, Nvidia, and AMD.

First we'll look at tests that use Compute shaders. Their appearance is one of the most important innovations in latest versions DX API, they are already used in modern games to perform various tasks: post-processing, simulations, etc. The first test shows an example of HDR rendering with tone mapping from the DirectX SDK, with post-processing using pixel and compute shaders.

The speed of calculations in compute and pixel shaders for all AMD and Nvidia boards is approximately the same, although video cards with GPUs of previous architectures had differences (curiously, the video card on Hawaii showed it again, albeit small). Judging by our previous tests, the results in the problem clearly depend not only on mathematical power and computational efficiency, but also on other factors such as memory bandwidth and ROP performance.

In this case, the speed of video cards is limited by bandwidth. Nvidia's new top-end board was 12% faster than its predecessor, the GTX Titan, in this test. If we compare the new product with the AMD board, then the Geforce GTX 780 Ti and the direct competitor Radeon R9 290X are approximately equal, although the Nvidia board is slightly more expensive.

The second compute shader test is also taken from the Microsoft DirectX SDK and shows the N-body gravity problem, a simulation of a dynamic particle system subject to physical forces such as gravity.

In the case of this test, the balance of power between the decisions of different companies turned out to be completely different. Nvidia graphics cards have a clear advantage in these calculation tasks, and Radeon graphics cards do not handle them very well. Therefore, it would be logical if this test was won by the most powerful of Nvidia’s boards - the Geforce GTX 780 Ti card presented today, which has more active computing units and operates at a high frequency.

But no, the GTX 780 Ti again lost a couple of percent to the more expensive GTX Titan in the computing task. Most likely, in calculation tasks, the frequency of the GK110 graphics processor in the case of a gaming video card drops below the level set in the case of the “computing” version - GTX Titan. As for the competitor, the Radeon R9 290X was left far behind, almost twice as much as the new Nvidia product.

Direct3D 11: Tessellation Performance

Compute shaders are very important, but another interesting innovation in Direct3D 11 is hardware tessellation. We looked at it in great detail in our theoretical article about the Nvidia GF100. Tessellation has been used for quite some time in DX11 games, such as STALKER: Call of Pripyat, DiRT 2, Aliens vs Predator, Metro Last Light, Civilization V, Crysis 3, Battlefield 3 and others. Some of them use tessellation for character models, others use it to simulate realistic water surfaces or landscapes.

There are several different schemes for partitioning graphic primitives (tessellation). For example, phong tessellation, PN triangles, Catmull-Clark subdivision. Thus, the PN Triangles partitioning scheme is used in STALKER: Call of Pripyat, and in Metro 2033 - Phong tessellation. These methods are relatively quickly and easily implemented into the game development process and existing engines, which is why they have become popular.

The first tessellation test will be the Detail Tessellation example from the ATI Radeon SDK. It implements not only tessellation, but also two different pixel-by-pixel processing techniques: simple normal map overlay and parallax occlusion mapping. Well, let's compare AMD and Nvidia DX11 solutions in different conditions:

In a simple bump mapping test, speed often comes down to bandwidth or ROP performance, and the result of the new Geforce GTX 780 Ti confirms this - it is almost identical to the speed of the GTX Titan in this test. All GeForces in this subtest are far behind the Radeon R9 290X, but not because of bandwidth, but because of the speed of ROP blocks.

In the second subtest, with noticeably more complex pixel-by-pixel calculations, things get a little more interesting. The efficiency of performing such mathematical calculations in pixel shaders is higher for GCN architecture chips than for Kepler, so it is not surprising that all Nvidia boards again lost out to the new solution based on the Hawaii chip. The Radeon R9 290X based on the new graphics processor is noticeably faster than the new GeForce GTX 780 Ti, which, in turn, overtook the GTX Titan by an impressive 18%, which approximately corresponds to the theory in terms of the speed of mathematical calculations.

In the tessellation test, the result of the new product is approximately the same as in the first subtest. The GTX 780 Ti model showed almost the same speed as the GTX Titan, losing to its direct rival in the Radeon R9 290X. This happened because in this tessellation test the division of triangles is moderate and the speed in it is not limited by the performance of the geometry processing units, so the triangle processing speed of AMD boards is enough to show good results.

The second tessellation performance test will be another example for 3D developers from the ATI Radeon SDK - PN Triangles. Actually, both examples are also included in the DX SDK, so we are sure that game developers create their code based on them. We tested this example with different tessellation factors to understand how much impact changing it has on overall performance.

But in this example, a more complex geometry is used, therefore, a comparison of the geometric power of various solutions for this test brings other conclusions. All modern solutions presented in the material cope well with light and medium geometric loads, showing high speed, but in difficult conditions Nvidia GPUs are still much more productive.

The Geforce GTX 780 Ti model announced today showed an abnormally low result compared to the GTX Titan on the same GK110 chip. And the lag of 15-20% at the three simplest levels of tessellation cannot be explained by anything, because the GTX 780 Ti is faster than Titan in all theoretical parameters (except for video memory). We are likely seeing the result of a software error in the form of unoptimized drivers. And only with the most complex tessellation does the new product take the lead, as it should.

And the comparison with the competitor in difficult conditions is positive for the new product, because it has more geometric blocks compared to Hawaii. Therefore, the GTX 780 Ti is much faster than the new generation AMD card, but only in difficult conditions, when the Radeon speed is seriously reduced, while the new Nvidia card remains quite high.

Let's take a look at the results of another test, the Nvidia Realistic Water Terrain demo, also known as Island. This demo uses tessellation and displacement mapping to render realistic-looking ocean surfaces and terrain.

The Island test is not a purely synthetic test for measuring exclusively geometric GPU performance, since it contains both complex pixel and compute shaders, and such a load is closer to real games that use all GPU blocks, and not just geometric ones, as in previous geometry tests. However, the main one still remains the load on the geometry processing units.

We tested the solutions at four different tessellation ratios—in this case, the setting is called Dynamic Tessellation LOD. If at the very first triangle splitting factor, when the speed is not limited by the performance of geometric blocks, the new top-end video card from AMD shows a fairly high result, trying to compete with GeForce, but it does not reach the level of the GTX 780 Ti even in this case. And with an increase in geometric work, Nvidia's new product leaps ahead even further.

Nvidia video cards are very fast in this test; the new Geforce GTX 780 Ti turned out to be 5-10% more productive than the more expensive GTX Titan, as it should be according to theory, unlike the previous test. The competitor still doesn’t have enough speed to compete with Nvidia cards, although in real games the load on geometric blocks is much less, and everything will be completely different there.

Conclusions on synthetic tests

The results of synthetic tests of the Geforce GTX 780 Ti video card, which has become the most powerful board in Nvidia's top series, as well as the results of other video card models produced by both manufacturers of discrete video chips showed that the new board is one of the most powerful solutions on the market, and it should successfully compete with others top-end boards, despite the rather high price.

The main thing we determined is that the new product is clearly faster than the GeForce GTX Titan in most tests, and this with a noticeable difference in price in favor of the GTX 780 Ti. For gaming, it's no surprise that Nvidia's new board is one of the more powerful offerings at the top end of the price range. With the exception of some tasks, the Nvidia model announced today performed well compared to the powerful Radeon R9 290X. Our set of synthetic tests showed that in terms of performance they will compete with each other in games, especially since Nvidia solutions traditionally perform better there than in “synthetics”.

The new GeForce GTX 780 Ti is clearly aimed at those enthusiasts who are not ready to compromise and plan to play current and future games at maximum settings in the highest resolutions, and are willing to pay a little more money for it than the competing Radeon R9 290X costs. Those who already wanted to buy a Geforce GTX Titan for gaming will be most happy, and those who have recently bought it least of all. After all, the new Nvidia model is cheaper, but will be even more productive in games. Let's move on to evaluating the real performance of the GTX 780 Ti in games in the next part of the article.