7870 consumption. Characteristics of the tested video cards

• GPU: Radeon HD 7870 (Pitcairn)
• Interface: PCI Express x16
• GPU operating frequency (ROPs): 1000 MHz (1000 MHz nominal)
• Memory frequency (physical (effective)): 1200 (4800) MHz (nominal - 1200 (4800) MHz)
• Memory exchange bus width: 256 bit
• The number of computing units in the GPU / the frequency of the blocks: 20/1000 MHz (nominal - 20/1000 MHz)
• Number of operations (ALU) in a block: 64
• Total number of operations (ALU): 1280
• Number of texture units: 80 (BLF/TLF/ANIS)
• Number of rasterization blocks (ROP): 32
• Dimensions: 255×100×33 mm (the last value is the maximum thickness of the video card)
• Textolite color: black
• Power consumption (peak 3D/2D/sleep): 178/57/3W
• Output jacks: 1×DVI (Dual-Link/VGA), 1×HDMI 1.4a, 2×Mini-DisplayPort 1.2
• Support for multiprocessing: CrossFire X (Hardware)

The card has 2048 MB of GDDR5 SDRAM placed in 8 chips on the front side of the PCB.

Recall that the card requires additional power, and two 6-pin connectors.

And once again it is worth specifically mentioning the possibility of a video card going into a deep “sleep”. The standard Windows power saving settings are designed to dim the monitor after a period of inactivity of the system, and many people around the world computers regularly enter this mode. The remaining components of the system unit do not change their mode of operation (except, of course, the central processor, which goes to sleep) - in particular, the video card continues to work in 2D mode and at the frequencies set for this mode. So, now the Radeon HD 7xxx, when the monitor is turned off, abruptly resets the operating frequencies, consuming only 3 watts, and at the same time the fan stops! This allows you to generally reduce the noise of the system unit and power consumption; in addition, the cooler does not absorb dust at this time. But even this is not enough. When operating in a CrossFire system with two or more 7xxx cards, as soon as the 3D operation ends (the player returns to 2D), only the first card remains in operation, and all the rest are exactly the same go to sleep, consuming 3 watts and turning off their fans. We really liked this feature of the new accelerators!

About the cooling system.

AMD Radeon HD 7870 GHz Edition 2048 MB 256-bit GDDR5 PCI-E
The device consists of two parts: the main radiator and the casing. The main heatsink is based on the evaporation chamber — this type of cooler was already used by AMD on HD 6970 cards. which is driven by a cylindrical fan fixed at one end of the casing. This version of the cooler is very efficient, and it cools the core and memory chips at the same time. The fan in 2D mode runs at very low speeds, and even in 3D the maximum spin that we saw was only 34% of the maximum. Even then, there was no noise.

We have researched temperature regime using the MSI Afterburner utility (author A. Nikolaychuk AKA Unwinder) and obtained the following results (at nominal frequencies and with strong overclocking):

First, it should be noted right away that the accelerator accelerates simply gorgeous! We got a 150 MHz core clock boost! And at the same time, the maximum temperature on the core rose by only a couple of degrees. Secondly, we see that, in general, the heating does not exceed 80 degrees, which is more than good for such accelerators. Thirdly, we notice that the cooler speed did not increase much after 6 hours under load. SO is great!

The video card arrived to us without packaging and kit, so we omit the bundle issue.

Installation and drivers

Test bench configuration:

• Computer based on Intel Core i7-975 (Socket 1366)
  • processor Intel Core i7-975 (3340 MHz);
  • Asus P6T Deluxe motherboard on Intel chipset x58;
  • RAM 6 GB DDR3 SDRAM Corsair 1600 MHz;
  • WD Caviar SE WD1600JD 160 GB SATA hard drive;
  • power supply unit Tagan TG900-BZ 900 W.
• operating system Windows 7 64-bit; DirectX11;
• monitor Dell 3007WFP (30″);
• drivers AMD versions Catalyst 12.3; Nvidia version 295.72

vsync is disabled.

Synthetic tests

The synthetic test packages we use can be downloaded here:

• D3D RightMark Beta 4 (1050) with a description at 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 .

As synthetic tests for DirectX 11, we used examples from the Microsoft and AMD SDKs, as well as the Nvidia demo program. The first is HDRToneMappingCS11.exe and NBodyGravityCS11.exe from the DirectX SDK (February 2010) .

We also took applications from both video chip manufacturers: Nvidia and AMD. From ATI Radeon The SDK samples DetailTessellation11 and PNTriangles11 were taken (they are also in the DirectX SDK). Additionally, Nvidia's demo program was used - Realistic Water Terrain, also known as Island11 (author - Timofey Cheblokov, a well-known 3D graphics specialist).

Synthetic tests were carried out on the following video cards:

• Radeon HD 7870 HD 7870)
• Radeon HD 7950 with standard parameters (hereinafter HD 7950)
• Radeon HD 7770 with standard parameters (hereinafter HD 7770)
• Radeon HD 6970 with standard parameters (hereinafter HD 6970)
• Geforce GTX 570 with standard parameters (hereinafter GTX 570)
• Geforce GTX 560 Ti with standard parameters (hereinafter GTX 560 Ti)

To compare the results of today's tested video card from the Radeon HD 7800 line, these models were chosen for the following reasons. Radeon HD 7950 and HD 7770 are taken as neighboring models from the current generation line - it will be interesting to see how the novelty is relative to them. The Radeon HD 6970 is taken as the older single-chip model from the previous generation, especially since the prices for "outdated" models are now quite profitable.

The selected solutions of the competing company Nvidia were taken because the Geforce GTX 570 now has a price close to the studied AMD product, and is its competitor at the moment. The GTX 560 Ti is a little cheaper, and it would be nice to compare the HD 7850 with it, but, unfortunately, at the time of our tests, this board was not yet available.

Direct3D 9: Pixel Shaders benchmarks

For some time now, we have stopped using our own texturing and filling (fillrate) test for 32-bit textures from RightMark of the first version, since most video cards in it currently show numbers that are far from theoretically possible and clearly incorrect in general. The test is too old. Next, we'll take a closer look at the texturing speed results by numbers from the 3DMark Vantage test, which yield quite realistic numbers.

The first group of pixel shaders that we are considering is very simple for modern video chips; it includes various versions of pixel programs of relatively low complexity: 1.1, 1.4 and 2.0 found in old games.

These benchmarks are too simple for modern GPUs and are mostly limited by either texturing performance or fillrate (not considering memory bandwidth). And therefore, they do not show all the capabilities of modern video chips, but they are still interesting from the point of view of taking into account outdated gaming applications, which are still missing.

So, judging by the comparison of the Radeon HD 7870 and HD 7950, the performance in these tests is most often limited by the fill rate. Although there is also an influence of the speed of texture modules, so AMD video cards became winners in this test, and the HD 7870 competes with might and main not with the HD 7950, but with the HD 6970. It is quite possible that in the case of the HD 7950 and HD 7770 there is a lack of optimizations in the driver, fixed in later versions, one of which we used in this comparison.

Compared to competing models Nvidia GeForce, AMD's new product is clearly faster than both solutions, and the GTX 570 and GTX 560 Ti, which showed close speed. Let's look at the results of more complex pixel programs of intermediate versions:

So this time it turned out that three Radeon cards at once turned out to be very close to each other (except for the HD 7770). The Cook-Torrance test is more computationally intensive, the difference in it roughly corresponds to the difference in the number of ALUs and their frequency, but also depends on the speed of the TMU. That's why given test better fit graphic solutions AMD companies and these video cards are ahead of both Geforce, although the difference is not too great. The new HD 7870 is even slightly faster than the HD 7950, which is in line with the theory if rendering depends on fillrate.

In the second test of procedural water rendering “Water”, which is more dependent on the speed of texturing, a dependent sampling from textures of large levels of nesting is used, and therefore video cards are ranked in it by texturing speed, adjusted for different TMU usage efficiency. And in this test, AMD solutions have an advantage over Geforce, this time even more ahead of them. And the new HD 7870 is slightly inferior to the younger model from the top series, which corresponds to the theory.

Direct3D 9: Pixel Shaders 2.0 tests

These tests of DirectX 9 pixel shaders are more complex than the previous ones, they are close to what we currently see in multiplatform games, and fall into two categories. Let's start with the simpler version 2.0 shaders:

• Parallax Mapping- a method of texture mapping familiar from most modern games, described in detail in the article.
• Frozen glass— a complex procedural texture of frozen glass with controlled parameters.

There are two variants of these shaders: one with a focus on mathematical calculations and one with a preference for fetching values ​​from textures. Consider mathematically intensive options that are more promising in terms of future applications:

These are universal tests, the performance of which depends both on the speed of the ALU units and on the speed of texturing; the overall balance of the chip is important in them, as well as the efficiency of the execution of computational programs. This is yet another test showing that AMD's architecture outperforms Nvidia's GPUs in legacy tasks. The performance of the new AMD graphics card in the Frozen Glass test is better than its predecessor from the top series and even better than the Radeon HD 7950. This seems to be another test that is highly dependent on the fillrate.

In the second Parallax Mapping test, the new HD 7870 video card, although it remains in the lead, is still not so much ahead of the youngest of the current top models. And although Nvidia's solutions feel a little better here, they still can't catch up with competing AMD boards. Let's consider the same tests in the modification with the preference of samples from textures to mathematical calculations:

For boards with GPUs manufactured by Nvidia, the situation has become quite a bit better, but still, the texturing speed of modern AMD chips is higher and it was not possible to catch up with them. Radeon video cards have defended their advantage in these tests, and both GeForces compete only with the HD 7770, lagging behind all three Radeon cards by more than high level. The HD 7870 copes very well with the tasks, outperforms the HD 7950 and is only a little behind the HD 6970, whose VLIW architecture is better suited for simple tasks.

But these were all outdated tasks, with an emphasis on texturing and fillrate. Next, we will look at the results of two more pixel shader tests - but this time version 3.0, the most difficult of our pixel shader tests for Direct3D 9. They are most indicative in terms of modern PC games, among which there are many multiplatform ones. The tests differ in that they heavily load both ALUs and texture units, both shader programs are complex and long, and include a large number of branches:

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

In our most difficult DX9 tests from the first version of the RightMark suite, Nvidia's graphics cards have previously been the leaders, in contrast to all previous tests in the tests from our review. But in the latest AMD architecture, its creators were able to solve all the shortcomings, and now solutions based on GCN architecture chips in PS 3.0 comparison showed their strength and instantly became leaders.

Tests are no longer limited by the performance of texture fetches, but most of all depend on the efficiency of shader code execution. Previously, the Radeon HD 6970 improved AMD's position in this test, increasing efficiency in the transition from the VLIW5 to VLIW4 architecture, and almost caught up with the Geforce GTX 570, then Tahiti and Cape Verde strengthened the result, and now Pitcairn has shown itself even better. The improvement in performance in complex computing is very noticeable when comparing old and new AMD boards, and the Radeon HD 7870 is well ahead of the HD 6970 and is about on par with the HD 7950.

So, we again saw excellent results for the Radeon HD 7870, a new model from AMD, which has always outperformed its direct competitor, almost always outperformed its top-end predecessor from the HD 6900 series, and was actually on par with the more expensive HD 7950. As for Nvidia's competitors , then the victory unconditionally goes to the new Radeon model, the presented novelty in all tests is noticeably faster and the GTX 560 Ti and GTX 570.

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

The second version of RightMark3D included two familiar PS 3.0 tests under 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 additionally increases the load on video chips.

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

The first pixel shader test will be Fur. At the lowest settings, it uses 15 to 30 texture samples from the heightmap and two samples from the main texture. The Effect detail - "High" mode 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 the maximum "severity" - 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.

The performance in this test depends mainly on the number and efficiency of TMUs, and on the efficiency of executing complex programs. In the variant without supersampling, the effective fillrate (ROP performance) and memory bandwidth also have an additional impact on performance, but to a lesser extent. The results when detailing the "High" level are up to one and a half times lower than when "Low".

In tests of procedural fur rendering with a large number of texture fetches, Nvidia solutions used to be noticeably stronger, but over the course of a couple of generations of GPUs, AMD has not only reduced the difference, but with the release of GCN, it has completely pulled ahead. And now the Radeon HD 7770 is on par with the GTX 560 Ti, not to mention the older models. The HD 7870 reviewed today is only slightly inferior to the older HD 7950 and showed a very good result, close to the best. This clearly indicates an increase in the efficiency of the new architecture in complex calculations.

Let's look at the result of the same test, but with "shader" supersampling turned on, which quadruples the work: maybe something will change in this situation, and memory bandwidth with fillrate will have less effect:

Enabling supersampling quadruples the theoretical load, and the results of Nvidia's solutions are significantly worse than those of AMD video cards. Now the difference in the efficiency of this task has become simply huge, and both tested Nvidia video cards simply lose to all AMD representatives. The new product from the HD 7800 series shows an excellent level of performance, again almost not losing to the HD 7950 and overtaking the HD 6970 by a margin.

The next DX10 test measures the performance of executing complex looping pixel shaders with a large number of texture fetches and is called Steep Parallax Mapping. At low settings, it uses 10 to 50 texture samples from the heightmap and three samples from the main textures. When you turn on heavy mode with self-shadowing, the number of samples is doubled, 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. We first check simple options without supersampling:

The second Direct3D 10 pixel shader test is more interesting for us from a practical point of view, since parallax mapping varieties are widely used in games, and heavy variants, like steep parallax mapping, are used in many projects, for example, in games of the Crysis and Lost Planet series. In addition, in our test, in addition to supersampling, you can turn on self-shadowing, which increases the load on the video chip by about two times - this mode is called "High".

The diagram is similar to the previous one without the inclusion of SSAA, Nvidia solutions could not improve their position. In the updated D3D10 version of the test without supersampling, the best of Geforce competes only with the Radeon HD 6970, and the new generation boards, except for the HD 7770, are far ahead. Thus, the HD 7870 almost did not yield to its older brother HD 7950 and became one of the two clear leaders in comparison. Let's see what will change the inclusion of supersampling, because it usually causes a large drop in performance on Nvidia boards.

When supersampling and self-shadowing are enabled, the task becomes even more difficult, the combined inclusion of two options at once increases the load on the cards by almost eight times, causing a serious drop in performance. The difference between the speed indicators of the tested video cards has changed, the inclusion of supersampling has an effect, as in the previous case - AMD cards have significantly improved relative performance compared to motherboards based on Nvidia chips.

This time, the Radeon HD 7770 again almost caught up with the Geforce GTX 570, and the two older video cards of the new AMD family became the best. The new Radeon HD 7870 is only slightly behind the stronger HD 7950, but only in simpler conditions, where the memory bandwidth affects. In general, according to the D3D10 shader tests reviewed, we can once again confirm the conclusion that the new AMD architecture copes well with complex "shader" tasks, much better than competing Nvidia boards from the previous generation.

Direct3D 10: PS 4.0 Pixel Shader Benchmarks (Computing)

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

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

The results of limiting mathematical tests most often correspond to the difference in frequencies and the number of computing units, but with the influence of different efficiency of their use. All AMD architectures over the past few years in such cases have had an overwhelming advantage over competing Nvidia video cards, and in the case of a comparison of GCN with Fermi, the situation has only slightly changed.

The results of the video cards are located on the diagram approximately in accordance with the theory, with a few exceptions. It is interesting that the leading trio of Radeons show results close to each other, which, by the way, confirm the theory - HD 7870, HD 7950 and HD 6970 have similar peak characteristics of ALU units. It is clear that all of them turned out to be much faster than both. Geforce cards, with only a little more than half that power.

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

And again, we see an almost identical to the previous picture, except that the numbers themselves have changed, but not their ratio. All graphic processors remained approximately on the same positions. Although there is no strict correspondence to the theoretical figures for peak performance, the results of all solutions are quite close to them. Thus, the difference between the HD 7770 and HD 7750 is close to the corresponding ratio of peak figures in terms of computing speed.

The diagram is fully consistent with the theory. The rendering speed in this test is limited solely by the performance of the shader units and their efficiency, so the three Radeon boards again showed close results, becoming the best comparison cards. It is clear that the HD 7770 is seriously inferior to them, as well as both Geforce - both Nvidia boards compete only with the weakest AMD. Actually, the conclusion is simple: before the release of Kepler, nothing threatens AMD in limiting computational problems, they still win all purely mathematical battles.

Direct3D 10: Geometry Shader Tests

There are two geometry shader speed tests in RightMark3D 2.0, the first option is called "Galaxy", the technique is 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 balance in the geometry shader tests does not affect the final rendering result, the final image is always exactly the same, only the scene processing methods change. The "GS load" parameter determines in which shader the calculations are performed - in vertex or geometry. The number of calculations is always the same.

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

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

It seems that the memory bandwidth has become the main limiter in this case - that is why the latest video card of the Radeon HD 7800 series seriously loses to its older sister HD 7950. However, it is ahead of the HD 6970 from the previous generation, so not everything depends only on memory. It is very interesting that the results of the HD 7900 and GTX 570 family cards almost completely correspond - they both are at the top and show almost one and a half times better performance in this test. Still, the truncated memory bus sometimes affects synthetic tests as well.

Let's see how the situation changes when transferring part of the calculations to the geometry shader:

When the load changed in this test, the numbers remained almost unchanged for Nvidia solutions and only slightly improved for newer AMD boards. All video cards in this test 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. Let's see what will change in the next test, which assumes a heavy load on geometry shaders.

"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 generation by drawing to two buffers, as well as new opportunity 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 rendering. For each point of the beam, 14 vertices are built in a circle, in total up to a million output points.

A new type of shader program is used to generate "rays", and with the "GS load" parameter set to "Heavy" - also to draw them. That is, in the "Balanced" mode, geometric shaders are used only to create and "grow" rays, the output is carried out using "instancing", and in the "Heavy" mode, the geometric shader is also involved in the output.

Unfortunately, due to a bug in the driver, this test simply did not run on the Radeon HD 7870. Alas, we could not run the most difficult geometry test, which shows all the capabilities of the GPU in processing geometry and the speed of execution of geometry shaders. And although we know that these capabilities are clearly improved in the new AMD chips, Radeon solutions continue to lag behind Geforce in the most difficult tests.

Direct3D 10: texture fetch rate from vertex shaders

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

Consider the first test "Earth", first in "Effect detail Low" mode:

Our previous research has shown that both texturing speed and memory bandwidth can affect the results of this test. And the results of Nvidia video cards in simple modes are limited by something else. And in general, in difficult conditions, the difference between boards of similar class turns out to be very small - percentages, not times.

So this time, except that the Radeon HD 7770 is far behind the rest of the set of video cards. All other solutions performed well. Moreover, AMD's new board from the Radeon HD 7800 family showed the best results in all modes, outperforming not only the predecessor Radeon HD 6970, but also the top-end HD 7950. Let's look at the performance in the same test with an increased number of texture fetches:

The relative position of the cards on the diagram has changed mainly due to the fact that the Nvidia boards provided high rendering speed in heavy modes. With a small number of polygons, the rendering speed for AMD cards is limited by the memory bandwidth, and Nvidia cards cannot show results better than the average - a clear emphasis on something (also memory bandwidth?). But in heavy modes, both Nvidia video cards improved their results and now compete with the Radeon HD 7950 and the new HD 7870, which are almost on a par.

Let's consider the results of the second test of texture fetches from vertex shaders. The Waves test has fewer 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 only slightly resemble what we saw in the previous diagrams. In this test, AMD and Nvidia video cards lined up almost along a clear ladder, with the exception of the Radeon HD 7770, which fell out of the trend. Except for this low-end AMD video card, all other motherboards from the company outperformed both Geforces.

And the best part is that the new model from the HD 7800 family showed the best result, again outperforming even the Radeon HD 7950 from the top family. It seems that there is no longer a performance focus in the memory bandwidth, and the HD 7870 wins due to better performance ROP blocks. Consider the second version of the same test:

And again there were changes similar to those that we saw earlier - all video cards slightly worsened their results, but Nvidia lost more, which allowed AMD-based boards to win an even clearer victory over them. Even the Radeon HD 7770 at heavy settings competes with the Geforce GTX 570. And today's heroine, the HD 7870, has again become the best board.

On the whole, the new board from the HD 7800 family showed itself quite well in vertex sampling tests, always outperforming not only competing solutions from Nvidia, but almost everywhere outperforming the previously announced board from the Radeon HD 7900 series, based on a more complex and expensive GPU. Excellent result!

3DMark Vantage: Feature tests

Synthetic tests from the 3DMark Vantage package will show us what we previously missed. Feature tests from this test suite support DirectX 10 and are interesting because they differ from ours and are still relevant. When analyzing the results of the new video card from the Radeon HD 7800 line in this package, we will draw some new and useful conclusions that have eluded us in the tests of the RightMark family.

The first test is the texture fetch speed test. Used to fill a rectangle with values ​​read from a small texture using multiple texture coordinates that change every frame.

Although the test by Futuremark does not show the theoretically possible level of texture fetching performance, the efficiency of AMD and Nvidia video cards in it is quite high and the comparative figures are close to the corresponding theoretical parameters. The top model of the Radeon HD 7000 family became the best video card in comparison, which confirms the theoretical results.

The Radeon HD 7870 reviewed today shows the second result, lagging behind only one of the top boards and ahead of the Radeon HD 6970, which is very, very good. All other boards are far behind. In the case of the Radeon HD 7770, this is due to its budget level, and in the case of both Geforces, this is due to the too small number of TMUs. Nvidia video cards are always weak in this test, even the best of the GTX 560 Ti pair shows a result far from the level of three similar Radeon video cards. The difference between models based on Pitcairn and Tahiti is fully consistent with the theory.

This 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 offscreen buffer (render target) using alpha blending. It uses a 16-bit FP16 off-screen buffer, the most commonly used in games that use HDR rendering, so this test is quite timely.

As you can see, the situation in the ROP performance test is quite different. As we determined earlier, the numbers of this subtest from 3DMark Vantage, although they show the performance of ROP units, but with a huge impact on the amount of video memory bandwidth (the so-called "effective fillrate"). This is clearly seen from the comparative results of Radeon HD 7000 family boards.

Alas, in this case, the test measures memory bandwidth rather than ROP performance, and therefore the Radeon HD 7870 is inferior not only to the HD 7950, but also to the HD 6970 - in full accordance with the theory. And the new model from AMD in this comparison showed a result comparable to the speed of the GTX 570, which confirms the theory about the speed limit of the memory bandwidth.

One of the most interesting feature tests, since this technique is already used in games. It draws one quadrilateral (more precisely, two triangles) using the special Parallax Occlusion Mapping technique, which imitates complex geometry. Rather 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 and heavy pixel shader for a video chip, containing numerous texture fetches during ray tracing, dynamic branching, and complex Strauss lighting calculations.

This test differs from the previous ones in that the results in it depend not only on the speed mathematical calculations, the efficiency of branch execution, or the speed of texture fetches, but a bit of everything. To achieve high speed, the right balance of the GPU is important here, and it also has a very noticeable effect on the speed and efficiency of branching in shaders.

And here the new family based on the latest GCN architecture proved to be just fine. The results of AMD's new graphics cards show that the Radeon HD 7000 series cards can handle the task very effectively in such complex computing tasks. Even the weak HD 7770 sits between the GTX 560 Ti and GTX 570, with all the other boards far ahead of them.

Comparative numbers of solutions based on AMD chips of different generations confirm the improved efficiency of executing complex calculations with branches on GPUs of the new architecture. Therefore, the HD 7870 was ahead of the HD 6970, although theoretically it has slightly less processing power. True, the HD 7950 is still ahead, but this is also easily explained by the specification of the video cards - the speed of the ALU units in the latter is 12% higher, the novelty is exactly the same and lagged behind the HD 7950 in the test.

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

The rendering speed in this test can also depend on several parameters, but the main influencing factors are geometry processing performance, geometry shader execution efficiency, and ROP performance. Therefore, it is quite logical that video cards manufactured by Nvidia, which have several geometric blocks, feel quite good in this application, and Geforce GTX 570 is ahead of all competitors, being the leader of the test.

But look - and the recently introduced model Radeon HD 7870 shows an excellent second in order result, ahead of the rest of the motherboards from AMD presented in the comparison! And this happens because of the higher fillrate and higher frequency of the GPU, on which the geometric blocks work. This is one of those tests that shows the advantage of Nvidia solutions with several geometric blocks, but AMD's new products with improved geometric performance perform well here.

A test for physical simulation of effects based on particle systems calculated using a video chip. Vertex simulation is also used, each vertex represents 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, their collisions with the height map are also calculated.

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

Another test from the 3DMark Vantage suite would have been similar to the previous chart, but geometry block performance is even more important. And that's why Radeon boards simply failed in relation to Geforce. See for yourself, now the GTX 570 has the best result, and the second one is the GTX 560 Ti.

Unfortunately, the GPU board codenamed Pitcairn doesn't really shine here, sitting between the Cayman and Tahiti Pro. Moreover, it is inferior to the HD 7950 rather because of the low fillrate and memory bandwidth, rather than the speed of geometric blocks. Well, even before the previous generation of Nvidia boards, the novelty is far away. So, nothing has changed in the synthetic tests of simulating fabrics and particles from the 3DMark Vantage test suite, which actively use geometry shaders - the new AMD solution is hampered by low memory bandwidth and fillrate (ROP unit performance).

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 increase the load 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, which shows the peak performance of video chips in extreme tasks, we see a slightly different distribution of results compared to similar tests from our Rightmark test package. This time, the performance of the solutions shown in the diagram corresponds to the theory only approximately, and somewhat differs from what we saw earlier in the mathematical tests from the RightMark 2.0 package.

Note that the new GCN architecture copes well with this task, clearly better than Nvidia's outdated Fermi. AMD solutions in general are much faster in such tests, and the youngest of the Geforce boards shows speed at the level of a card from the Radeon HD 7700 line, and even the GTX 570 cannot catch up with any of the top three fastest Radeons. AMD graphics cards always show top scores in cases where simple and intensive mathematics is performed.

As for the recently introduced novelty, compared to its relatives, we note that it outperformed the Radeon HD 6970 quite confidently and only slightly outperformed the board from the older HD 7950 line. This is a very good result, considering that, theoretically, it should be 12 %, and the difference in practice was only 2%.

Direct3D 11: Compute Shaders

To test the new solutions from AMD and in tasks that use new DirectX 11 features such as tessellation and compute shaders, we used examples from the developer kits (SDKs) and demo programs from Microsoft, Nvidia and AMD. Unfortunately, we have not yet tested all the solutions in these tasks, and we will have to compare the HD 7870 model not with the GTX 560 Ti and GTX 570, but with the GTX 580, which is not entirely correct. Yes, and instead of the HD 7950, you will have to consider the top-end HD 7970.

First, we'll look at benchmarks that use Compute shaders. Their appearance is one of the most important innovations in the latest versions of the 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 , which uses pixel and compute shaders.

Although this is not the most successful example for compute shaders, it still shows the difference in performance. There is almost no difference in the speed of calculations in the compute and pixel shaders for AMD and Nvidia video cards, but the former are slightly faster in the compute shader, and Nvidia in the pixel shader.

It is possible that the results clearly depend not only on mathematical power, but also on memory bandwidth. And yet, the lag looks quite plausible. new Radeon HD 7870 outperforms HD 7970. Interestingly, the new product is also inferior to HD 6970, which is faster in mathematics and has a larger memory bandwidth. It is clear that the top model HD 7970 is the leader, and the GTX 580, although slightly, is still faster than the HD 7870.

The second compute shader test is also taken from the Microsoft DirectX SDK and shows N-body gravity computational problem - simulation dynamic system particles that are subject to physical forces such as gravity.

The results in this test are also very similar to what we saw in the previous test, except that the HD 7870 and HD 6970 are swapped. That is, this test rather measures the speed of mathematical calculations. New AMD model almost caught up with the more expensive Geforce GTX 580, which cannot be considered its direct competitor, since it costs more. In general, the result of the novelty looks good, if we take into account the not very big gap from the best single-chip video card in general. It will be interesting to see the performance in the tessellation tasks we are now moving on to.

Direct3D 11: Tessellation Performance

Compute shaders are very important, but another important innovation in Direct3D 11 is hardware tessellation. We considered it in great detail in our theoretical article about Nvidia GF100. Tessellation has been used in DX11 games for a long time, such as STALKER: Call of Pripyat, DiRT 2, Aliens vs Predator, Metro 2033, Civilization V, Crysis 2, Battlefield 3 and others. Some of them use tessellation for character models, others to simulate a realistic water surface or landscape.

There are several different schemes for partitioning graphic primitives (tessellation). For example, phong tessellation, PN triangles, Catmull-Clark subdivision. So, the PN Triangles tiling scheme is used in STALKER: Call of Pripyat, and in Metro 2033 - Phong tessellation. These methods are relatively quick and easy to implement 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: a simple overlay of normal maps and parallax occlusion mapping. Well, let's compare DX11 solutions from AMD and Nvidia in different conditions:

We have already seen that parallax occlusion mapping (middle columns in the diagram) on video cards from both manufacturers is much less efficient than tessellation (lower columns), and tessellation does not lead to a significant drop in performance - compare the upper and lower columns. That is, high-quality geometry simulation using pixel calculations provides even lower performance than tessellated geometry with displacement mapping.

As for the performance of video cards relative to each other, there is something to think about. In the simple bumpmapping test, you can see that the boards most likely hit the memory bandwidth again, since their results are too close. In all other respects, the picture is more or less plausible, AMD motherboards are ranked and the top model is in the lead. The Radeon HD 7870 is close to the GTX 580 and HD 6970 in this subtest.

But the second subtest with complex pixel calculations shows that the efficiency of performing complex mathematical calculations for GCN architecture chips is much higher than for other participants in the comparison. Thus, the board of the HD 7800 family based on the Pitcairn chip showed a very good result in the parallax mapping test compared to the HD 6970 and GTX 580, which again indicates a very high efficiency of executing complex shader programs on the Southern Islands chips.

In the most interesting tessellation subtest, the triangle splitting is quite moderate, and therefore AMD's boards don't lose much performance. There is also enough headroom to get ahead of the fastest single-chip Nvidia graphics card. But the most interesting conclusion of this test is that the Radeon HD 7870 was faster than the Radeon HD 7970 here! This could be explained by the higher frequency of the first GPU (after all, it even has a special “GHz Edition” index), on which the geometric blocks work.

But after all, the difference in frequency is even less than the difference in the performance of this test, and we found out that not only the geometry processing speed is important in it. More like some more aggressive driver optimizations. However, in any case, the board at Pitcairn showed a very decent result for a board in this price range. Let's check the tessellation next.

The second tessellation performance test will be another example for 3D developers from the ATI Radeon SDK - PN Triangles. In fact, both examples are also included in the DX SDK, so we are sure that game developers create their own code based on them. We tested this example with a different tessellation factor to see how much it affects the overall performance.

An attentive reader will definitely note that this time we simply do not have the results with the maximum tessellation level (tessellation factor = 19) that we indicated earlier on the diagram. Even when testing the Radeon HD 7700 line, we encountered the fact that the test does not allow to set maximum value this setting. Then we assumed that this AMD video driver "cuts" the capabilities so that the company's video cards do not look so bad in synthetic tests of extreme tessellation, but it turned out that it was not the drivers at all...

Our research showed that changes to disable the maximum tessellation level were made to the source code and the compiled example from a newer version of the DirectX SDK. If the PNTriangles11 example in the February 2010 version of the DX SDK still allows you to set the Tess factor to 19, then a change was made to the June SDK of the same year that prohibits the tessellation level above 9 (all this is confirmed by the source code of the example from the SDK, one of the constants in the new its version has been replaced with a smaller value that does not allow setting a parameter that was possible in the previous version).

Who did it and why? Is fecit, qui prodest. We are looking for who benefits by considering the facts. Fact one: PNTriangles11 is an example of tessellation implemented by AMD and included in both their own SDK and the DirectX SDK. Fact two: AMD video chips cope with tessellation worse than competing ones at maximum levels of triangle splitting. The conclusion is very simple: it looks like AMD asked Microsoft to change the source code for this example in the June 2010 DX SDK with a minor fix that has little to no effect on its functionality. Nearly.

We understand that such levels will not be used in games. We also understand that the triangles are too small to be effective. But in our case, this is a purely synthetic test that allows us to find out the theoretical limits of the hardware's capabilities. Let fair results always be shown in synthetics, and we will tell you that there will not be such complex geometry in games for a very long time. Extreme partitioning factors are unlikely to be used in games in the near future, but in this section we are interested in the architectural potential. In general, such a fix is ​​not the most beautiful move on the part of AMD, as it seems to us.

Well, let's consider at least three levels of tessellation, without a maximum. In this example, we see a more plausible comparison of the geometric power of various AMD and Nvidia solutions. All modern chips cope well with such a load (without tess factor = 19), and here we can only distinguish the Radeon HD 7770, which is inferior to the rest due to general weakness. But in general, all the new AMD GCN architecture chips are very good, and the Radeon HD 7970 even outperformed the Geforce GTX 580.

As for our today's hero, the Radeon HD 7870 is only slightly inferior to its older brother in the face of the top motherboard of the current generation, which is simply an excellent result! It can be seen that the chips of the GCN architecture in tessellation are noticeably faster than the Cayman chip, even the budget one. Therefore, we can expect Pitcairn solutions to be strong in other existing tests using tessellation, such as 3DMark 11 and Heaven.

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

Island is not a purely synthetic test for measuring geometric performance alone, it contains both complex pixel and compute shaders, and such a load is closer to real games that use all GPU units at once, and not just geometric ones, as in the previous benchmark .

We tested the program with four different coefficients tessellation, in its case the setting is called Dynamic Tessellation LOD. And if AMD video cards are very strong at low triangulation factors, then when the work becomes more complicated, the only Nvidia board starts to win. With an increase in the split ratio and scene complexity, the performance of all Radeons falls more strongly, and here it would be possible to recognize Nvidia's victory in complex geometric tests again and leave it at that.

But that's if you don't pay attention to the obviously anomalous result of the Radeon HD 7870 we are considering today. Somehow it happened that it overtook the HD 7970, and even with a huge advantage! This simply cannot be in principle, based on theoretical specifications, so again you need to figure it out. This time, special optimizations in the latest version of the AMD Catalyst video driver are to blame. We tested the HD 7970 and HD 7770 with previous versions of drivers that did not have these optimizations, but they appeared in the latest Catalyst. And now the Radeon HD 7870 is almost at the level of the Geforce GTX 580. Where the HD 7970 lost a lot. Story? No, rather - well-known tessellation optimizations, which were previously noted only in gaming applications, but not in pure synthetics. Alas, we again note some cunning on the part of AMD.

If we do not take this into account, we note that under conditions of a heavy geometric load, the new chips are still very good, the number and efficiency of geometric blocks in different chips of the family is the same, and the difference in speed is due to different clock frequencies and the impact on the overall performance of other parameters. In general, GCNs have greatly improved geometric performance and in real applications they are not inferior to Nvidia's Fermi. I wish AMD didn't use dubious "optimization" methods...

Conclusions on synthetic tests

Based on the results of synthetic tests of a new video card model from the Radeon HD 7800 series based on the Pitcairn graphics processor from the Southern Islands family, as well as the results of other video card models produced by both manufacturers of discrete video chips, we conclude that new mid-budget solutions should fit well into the line AMD and become one of the most profitable purchases. The Radeon HD 7870 looks very good in synthetic tests, often performing at the level of the youngest of the top Tahiti-based video cards, but at the same time it costs noticeably less and will be sold more massively.

The Pitcairn GPU is built using the latest 28nm process technology based on the new GCN architecture, which is very different from the company's previous solutions. Note that the new family of GPUs has a lot of architectural improvements aimed at increasing the efficiency of performing complex calculations on the GPU and speeding up the processing of geometric data (including tessellation). Our set of synthetic tests showed that the efficiency of computations in such problems has indeed increased.

True, there were some unpleasant discoveries here - since AMD chips cope with extreme degrees of tessellation worse than their rivals, the company tries by all means to show them in these tasks better than they really are. At the same time, sometimes very dubious methods are used, the details of which can be read a little higher. It is not very clear why this is being done, since the lag is observed only in purely synthetic applications, and in games everything is fine with the new chips anyway.

In general, thanks to the architectural changes made, the video cards of the new series should become very profitable option for the purchase and modernization of gaming systems, they will definitely be one of the best-selling DirectX 11 solutions, and in the future will improve AMD's position in the market, especially given the lack of 28-nanometer competitor's new products. Moreover, unlike the junior solutions on Cape Verde, video cards on Pitcairn do not suffer from low video memory bandwidth, are perfectly balanced and show very strong results.

In fact, the Radeon HD 7800 family is the middle ground that will suit most gamers best - they are not too expensive like the HD 7900, but not too slow like the HD 7700. We are confident that the strong performance of the Radeon HD 7870 in synthetic tests will be supported by excellent performance in gaming applications. New boards are required to show competitive speed in games compared to Nvidia rivals, and in the next part of the material we will just check this.