Why do pixels burn out on an LCD TV? Burn-in of OLED TVs in real tests

AMD Radeon HD 6950/6970:

description of video cards and results of synthetic tests

It makes sense to remind you once again that the cards require additional power, and the 6950 requires two 6-pin connectors. And 6970 - 8-pin and 6-pin. We hope that AMD's partners will include the appropriate power splitter adapters in the kit.

About the cooling system.

AMD Radeon HD 6950/6970 2048 MB 256-bit GDDR5, PCI-E

It's worth noting that the CO is very similar in principle to what we saw on the GTX 580/570, and it is also based on a vapor chamber, which is contained in a narrow copper compartment in contact with the GPU. Above this chamber there is a structure of cooling fins through which air passes, driven by a cylindrical fan at the end of the entire device. True, unlike the GTX 580, in this case the entire structure is made of copper, including the radiator fins, so the CO turned out to be very heavy. We have already written that such a solution is more effective than the one traditionally used previously on heat pipes. Inside the evaporation chamber there is a special liquid that instantly transfers heat from the bottom plate to the top. It is also especially worth noting that the CO is configured for minor reactions when heating to ensure almost silent operation. Therefore, core heating may even exceed what we saw in the case of the 5870.

We conducted a temperature study using the MSI Afterburner utility (author A. Nikolaychuk AKA Unwinder) and obtained the following results:

AMD Radeon HD 6970 2048 MB 256-bit GDDR5, PCI-E

AMD Radeon HD 6950 2048 MB 256-bit GDDR5, PCI-E

The results of the study showed that, despite all of the above, CO is really effective, and even at a rotation speed of 40% of the maximum, the heating is 92 degrees for the 6970, and 84 for the 6950. This is after 6 hours of constant testing under load in 3D. Yes, 92 degrees may seem excessively high to some, but for Hi-End accelerators this is acceptable.

The maximum power consumption of cards under load is 250-260 W for the 6970 and slightly higher than 205 W for the 6950. We deliberately do not provide any consumption graphs so as not to complicate the reading of the material. Readers are always interested in how much it consumes at maximum in order to select the right power supply, but few people are interested in the details.

Equipment. Considering that reference samples never have a complete set, we will omit this question.

Installation and drivers

Test bench configuration:

• Computer based on CPU Intel Core i7-975 (Socket 1366)
  • Intel Core i7-975 processor (3340 MHz);
  • Asus P6T Deluxe motherboard based on Intel X58 chipset;
  • RAM 6 GB DDR3 SDRAM Corsair 1600 MHz;
  • hard drive WD Caviar SE WD1600JD 160 GB SATA;
  • power supply Tagan TG900-BZ 900 W.
• operating system Windows 7 64-bit; DirectX 11;
• Dell 3007WFP monitor (30″);
• ATI drivers version Catalyst 10.11; Nvidia version 263.09 / 260.99.

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 a brief description: for Vista without SP1, for Vista with SP1.

In the absence of our own synthetic DirectX 11 tests, we once again used examples from the Microsoft and AMD SDKs and the Nvidia demo program. First, there are HDRToneMappingCS11.exe and NBodyGravityCS11.exe from the DirectX SDK (February 2010).

We also took applications from both manufacturers: Nvidia and AMD. The examples DetailTessellation11 and PNTriangles11 were taken from the ATI Radeon SDK (they are also in the DirectX SDK). Additionally, we used a demo program from Nvidia - Realistic Water Terrain, also known as Island11 (author - Timofey Cheblokov, a very famous specialist in 3D graphics).

Synthetic tests were carried out on the following video cards:

• Radeon HD 6970 HD 6970)
• Radeon HD 6950 with standard parameters (further HD 6950)
• Radeon HD 6870 with standard parameters (further HD 6870)
• Radeon HD 5870 with standard parameters (further HD 5870)
• GeForce GTX 580 with standard parameters (further GTX 580)
• GeForce GTX 570 with standard parameters (further GTX 570)

To compare the results of the new models of Radeon HD 6900 series video cards, these models were chosen because the Radeon HD 5870 is the company’s previous single-chip solution for the top price range, the strongest before the release of new models; Radeon HD 6870 is AMD's current solution, one step below the top ones and based on the recently released Barts video chip.

Namely, these Nvidia solutions were taken because the Geforce GTX 580 is the company’s fastest single-chip model, based on a fresh GPU. Although it is not a competitor to the presented video cards in terms of price, its results are interesting as a kind of maximum bar for Nvidia solutions. Well, the GTX 570 is taken as a direct competitor to the older model of the new series - HD 6970.

Direct3D 9: Pixel Filling tests

This test determines the peak texture sampling performance (texel rate) in FFP mode for different numbers of textures applied to one pixel:

In this test of filtering 32-bit (8 bits per color) textures, most video cards show numbers that are far from theoretically possible. So the results of our texture synthetics in the case of HD 6900 series video cards do not reach the peak values. Next, we'll look at texturing speed again, in a test from the 3DMark Vantage package, where we get more realistic numbers.

And here it turns out that the HD 6970 selects only 67 texels per clock cycle from 32-bit textures during bilinear filtering, which is almost a third lower than the theoretical figure of 96 filtered texels. For the HD 6950, these figures correspond to 62 texels out of 88 theoretical ones, that is, the efficiency of the younger model turned out to be slightly higher, and this is due to a slight difference in video memory bandwidth, which also affects the results.

It is not surprising that all AMD cards show such high performance and are significantly ahead of their rivals from Nvidia. After all, their theoretical texturing speeds are very high. But even the top-end GTX 580 has only 64 TMU and is much inferior to models on the Cayman, which have 88-96 TMU, and even operate at higher frequencies.

The difference between the HD 6950 and HD 5870 in different conditions turned out to be very interesting. If in cases with a large number of textures, where the number of TMUs and their frequency have the greatest impact, they are on par, then with a smaller number of textures per pixel, the HD 5870 model comes out ahead. Moreover, the difference cannot be attributed only to the memory bandwidth, and probably various factors also have an impact here driver optimizations.

Let's look at the same results in the fill rate test:

These numbers show the fill rate, and in them we see everything the same, except that we take into account the number of pixels written to the frame buffer. The maximum result remains with the new top solutions of the Radeon HD 6900 family, which have simply a huge number of TMUs and are more efficient in our synthetic test. Surprisingly, in cases with 0-4 overlaid textures, the youngest of the video cards we are considering today is for some reason much inferior to the previous top-end AMD solution, although in difficult conditions it practically does not lag behind it.

Direct3D 9: Pixel Shaders tests

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 older games.

The tests are very simple for modern GPUs and heavily focus on texturing performance. Therefore, they do not show all the capabilities of modern video chips, but they are still interesting for assessing the balance between texture samples and mathematical calculations. In this case, there are no special differences between the HD 5870 and HD 6950; the results of these models are comparable. Although one test stood out - the pixel shader of lighting with three sources according to Phong clearly depends on the mathematical performance of the GPU, and therefore only the older model, the HD 6970, reached the HD 5870 level in it.

Performance in other tests is limited mostly by texture unit speed and fill rate, but takes into account block efficiency and data caching. The new Radeon HD 6900 series models are slightly faster than their predecessors: the HD 6970 is faster than the HD 5870, and the HD 6950 is faster than the HD 6870 (from a different price range). And almost all of them are ahead of both top GeForce models - even the GTX 580 in these tests shows results only at the level of the HD 6870, and the lack of texturing speed is clearly to blame for this.

Let's look at the results of more complex intermediate pixel programs:

And this time it happened about the same, again the GTX 580 competes more with the HD 6870 than with the real top AMD models. The Cook-Torrance test is more computationally intensive, and the difference in it roughly corresponds to the difference in the number of ALUs and their frequency. That is why this test is better suited for the AMD architecture, whose chips have a larger number of mathematical units.

And here two interesting points were found. Firstly, the HD 5870 outperforms even the HD 6970, which is difficult to explain by theoretical characteristics alone. There is almost no difference in peak mathematical performance between these models, but there are also architectural differences. It seems that it was the different efficiency of execution of this shader on those same VLIW5 and VLIW4 processors that led to such a difference not in favor of the new Cayman chip. Therefore, the HD 6950 in this test performed only at the level of the HD 6870, as well as the GTX 580.

The second test of procedural water rendering, “Water,” which is highly dependent on texturing speed, uses dependent sampling from textures of large levels of nesting, and video cards in it are ranked by texturing speed, adjusted for different efficiency of TMU use.

In this test, the new solutions are doing great, the HD 6950 provides results at the level of the HD 5870, and the HD 6970 leads with a good margin, almost corresponding to a 25 percent difference in the theoretical texturing speed. It is clear that Nvidia video cards have nothing to gain here, and they show results at the level of a noticeably cheaper competitor model.

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 that depend on both the speed of the ALU units and the texturing speed; the overall balance of the chip is important in them. The performance of the new AMD graphics cards in the Frozen Glass test is quite good, the HD 6970 was again noticeably faster than the HD 5870, and the HD 6950 almost caught up. Unfortunately for Nvidia, due to weak texturing, AMD's solutions were again noticeably faster.

In the second “Parallax Mapping” test, Nvidia’s solutions already feel a little better, and the HD 6870 and HD 6950 are close to the results of the GTX 580 card from another market segment, which costs more. Interestingly, the HD 5870 was again faster than the HD 6970. This confirms our theory that the speed in the test is limited by mathematical performance and that the test is slightly less suitable for AMD's new architecture.

There is another probable explanation - synthetic tests often heavily load the GPU with parallel calculations, and the power consumption of new models in synthetics may well exceed the set limit. Consequently, the clock frequency may also decrease, and with it the results may be lower than expected. However, this assumption needs to be checked. Let's consider these same tests in a modification with a preference for samples from textures over mathematical calculations:

For Nvidia solutions, the situation has become noticeably sadder, since the texturing speed of the latest AMD chips, unlike their competitors, is very good, so they are only increasing their already undeniable advantage. Even the very best GTX 580 is inferior to the same HD 6870 in both tests with an emphasis on texturing. Well, our new heroes from the HD 6900 family turned out to be the fastest, the HD 6950 even beat the HD 5870, albeit by mere pennies. And the HD 6970 again became the leader, which is understandable theoretically if you look at the performance of TMU units.

These were all legacy tasks, mostly with an emphasis on texturing, and less often on fillrate. 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 the Direct3D 9 API. They are most indicative from the point of view of modern games on PC, many of which are multi-platform. 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.

In our most challenging DX9 tests, Nvidia graphics cards always perform better than AMD solutions, contrary to all previous tests. This situation is due to the fact that these tests are not limited by the performance of texture samples, but rather depend on the efficiency of pixel shader code execution.

In tests of complex pixel shaders version 3.0, the new top AMD video cards still could not catch up with their competitors, although they came noticeably closer to them. The speed in both PS 3.0 tests is slightly dependent on memory bandwidth and texturing, but the code is complex, which the new Nvidia architecture and... the new AMD architecture handle very well. Perhaps this is the first test where we see a noticeable positive difference between the previous and the latest AMD architectures.

And the latter copes with the task clearly better. Although even the HD 6970 can hardly compete with the GTX 570, we never even thought about this before. Nvidia solutions have always been the undisputed leaders in this pair of test tasks, and have traditionally shown much stronger results. And video cards based on the new Cayman graphics chip were able to come close to them.

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

The second version of RightMark3D included two 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, but varies across different conditions. The results at the “High” level of detail are approximately one and a half times lower than at the “Low” level, as it should be according to theory. In D3D10 tests of procedural fur rendering with a large number of texture samples, Nvidia solutions used to be noticeably stronger, but the latest AMD solutions have caught up to them, which we have already seen before.

In the non-supersampling option, the effective fill rate (ROP performance) and memory bandwidth have a greater impact on performance. Therefore, Nvidia’s solutions were ahead, and only the top-end Radeon HD 6970, presented today, almost catches up with the junior GTX 570. A lower-level model called HD 6950 shows results at the level of HD 5870, but the HD 6870 achieved approximately the same result. This is not surprising, because Its fill rate is even higher than that of older solutions of the HD 6900 series.

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:

As always, enabling supersampling increases the theoretical load by four times, and the results of Nvidia solutions drop noticeably compared to those of AMD video cards. Now the three models with similar results (HD 6870, HD 5870 and HD 6950) are ahead of the GTX 570, and the older solution HD 6970 successfully competes with the GTX 580. The difference between the top cards of the HD 6000 and HD 5000 lines remains approximately the same, the new model wins a few percent from the previous one.

The second DX10 shader 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 somewhat more interesting from a practical point of view, since varieties of parallax mapping are widely used in games, and heavy options like our steep parallax mapping are used in many projects, for example in the games Crysis and Lost Planet. 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 in many ways similar to the previous one (without SSAA), only Nvidia’s position has weakened somewhat. In the updated D3D10 version of the test without supersampling, the HD 6970 becomes on par with the GTX 570, which is normal for direct competitors, and the top GTX 580 remains the leader. The other three video cards from AMD show similar results and lag behind. Let's see what difference enabling supersampling will make; it can cause a significant drop in speed on Nvidia boards.

When supersampling and self-shadowing are enabled, the task becomes even more difficult; enabling both options together increases the load on the cards by almost eight times, causing a large 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 slightly improved their performance compared to Nvidia solutions.

Now the HD 6970 shows results at the level of the GTX 580, and the HD 6950 and HD 5870, which are approximately equal in speed, are on the same level as the GTX 570. And only the cheaper HD 6870 is slightly behind this Nvidia video card. The comparative figures for the HD 6970 and HD 5870 pairs were repeated again, the difference in favor of the more recent models is about the same. Based on these tests, we can conclude that both cards of the HD 6900 family released today coped with shader tasks very well, at the level of Nvidia’s traditionally strong competitors in these tasks.

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 the extreme mathematical tests usually correspond to the difference in the frequencies and number of execution units, but with the influence of their efficiency. Modern AMD architecture has a big advantage over competing Nvidia graphics cards in such cases, and this explains the test results, in which AMD solutions clearly turn out to be significantly more productive, although not as great as their theoretical advantage.

Theoretically, the GTX 580 should be almost twice as slow as the HD 5870 and HD 6970. In practice, the difference is not even one and a half times. Of course, this doesn’t change much, because even the HD 6870 is significantly faster than both Nvidia cards in such tests, not to mention the top models. Otherwise, the solutions fell roughly in line with theory, with a few exceptions.

For example, the results of comparing the new and old top families of AMD video cards turned out to be interesting. Firstly, the HD 6870 performed identically to the HD 6950 in this test, with the difference in theoretical figures favoring the Cayman-based model. Secondly, the same can be said about the combination of HD 6970 and HD 5870 - with similar theoretical figures, in reality the older one, with stream processors based on the VLIW5 architecture, wins by a small margin.

Here again, there are several possible explanations - either AMD has not yet fully optimized the drivers for the new GPUs, or the Cayman architecture is less efficient in this test (while it is quite possible that it will be more effective in less straightforward tests), or PowerTune technology has influenced In this test, the limitation of video memory bandwidth also began to take its toll.

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:

This time, all GPUs remained in approximately the same positions, except for the relative performance of the Cayman and Cypress/Barts. Now in these pairs everything is in strict accordance with the theoretical peak performance figures, and the HD 6970 is even slightly ahead of the HD 5870, that is, in this case the new architecture worked more efficiently. And in the pair HD 6950 and HD 6870 there is now such a difference in favor of the top solution, as it should be.

The rest is nothing new. Since the rendering speed here is limited solely by the performance of the shader units, the HD 6970 and HD 5870 are the leaders, followed by the rest of the AMD video cards, and both GeForces are inferior, including the younger model from a different price range. Although the advantage of AMD solutions still remains somewhat lower than when comparing theoretical figures - this suggests that the efficiency of superscalar processors VLIW5 and VLIW4 is below 100%.

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 drops by about two times. The task for modern video cards is not particularly difficult; performance in general is limited not only by the speed of geometry processing, but also by memory bandwidth or fill rate to a certain extent (within one manufacturer).

Alas, although we previously saw an increase in the geometric performance of solutions on Barts in this test, this time the video cards of the new family turned out to be approximately at the same level as the Radeon HD 5870 of the previous generation. Perhaps the performance limitation of the video memory is to blame, but the HD 6870 is very strong in this test and even outperformed the HD 6950. So most likely the effective fill rate, that is, the ROP performance, is to blame.

In any case, all AMD solutions are very far from top-end Nvidia video cards, and although the execution of geometry shaders may have become more efficient, this is clearly not enough. Nvidia graphics cards based on the GF110 perform almost twice as fast as all competitor graphics cards. Let's see how the situation changes when we transfer part of the calculations to the geometry shader:

When the load changed in this test, the numbers remained almost unchanged for both Nvidia and AMD solutions. In this test, new video cards of the HD 6900 family react weakly to changes in the GS load parameter, which is responsible for transferring part of the calculations to the geometry shader, like other solutions, but still show results slightly higher than in the previous diagram. Let's see what changes in the next test, which involves 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 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. That is, 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. First we look at the easy mode:

The relative results in different modes again roughly 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 less than twice as slow.

In this test, rendering speed should be limited by geometric performance, but the primitives processed are clearly not enough for AMD's new architecture to perform significantly better, although there is a small difference that is explained by architectural changes in the GPU.

Nvidia video cards still remain the leaders in the test, but the same Radeon HD 6970 has almost caught up with the younger model GTX 570. And the HD 6950 outperforms the HD 5870, although not too much. And these good results clearly indicate the presence of optimizations for processing geometric data in new chips.

The numbers should change 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.

But in this test, the difference between AMD chips with a traditional graphics pipeline (including the Cayman with its two rasterizers) and chips with the Fermi architecture is immediately noticeable. Although we know from previous studies that low-end Nvidia chips lag behind in terms of execution speed of geometry shaders, showing less impressive results, since their geometric processing capabilities are reduced. But the results of the GTX 570 and GTX 580, based on the GF110 chip, are very good and almost twice as high as those of the best AMD solution.

And this solution is the brand new Radeon HD 6970. The capabilities of the new top chip for processing geometry and the speed of execution of geometry shaders have clearly increased compared to other video cards from the company. And new solutions based on Cayman show results in these tests higher than solutions based on Cypress and Barts, although not three times, or even twice. Probably, AMD engineers still have to solve the problem of parallelizing the work of triangle installation blocks (geometry setup), which these tests may run into.

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 are affected by both texturing speed and memory bandwidth. This is clearly seen in the comparative results of the Radeon HD 5870 and HD 6950, and other AMD solutions. It seems that it is the memory bandwidth that limits their performance in the test, so the difference between all the solutions is not that great.

Nevertheless, the HD 6970 shows very good results on the new GPU - it almost matches the GTX 570, with which this model will have to compete in the real world. Well, the leader remains the most expensive and productive GTX 580. Both cards of the HD 6900 family performed well, the youngest new model is almost on par with the previous top-end one. Let's look at the performance in the same test with an increased number of texture samples:

The relative position of the cards on the diagram has changed noticeably, especially in hard mode. Although Nvidia video cards for some reason lost performance precisely in the easiest conditions. Just with a small number of polygons, the speed is limited by memory bandwidth, and in this case, the new AMD boards have almost caught up with the competitor’s top solutions.

But in heavy modes, the difference in favor of Nvidia has grown to one and a half times, where the GTX 580 and GTX 570 remain out of reach of their rivals. The senior video card of the HD 6900 family outperforms other AMD solutions, although this is again barely noticeable when compared with the HD 5870. One could talk about the influence of memory bandwidth, but this does not stop the competitor...

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.

Interestingly, the results in the second "Waves" vertex texturing test are not at all similar to what we saw in the previous charts. In this test, all AMD and Nvidia video cards show very close results, which can also be attributed to the limitation of video memory bandwidth. This figure for all presented video cards is in the region of 130-190 GB/s, and the spread is small. The latest model Radeon HD 6970 again became the best among AMD video cards. Let's consider the second version of the same test:

And again, changes occurred similar to what we saw earlier - Nvidia video cards “sagged” only in easy mode, and AMD in all three. And therefore, in the mode with a small number of polygons, the difference between the solutions is small, but in medium and heavy modes, the GTX 580 and GTX 570 are noticeably ahead of all Radeon models, including those from the HD 6900 family announced today. Compared to Cypress, the new GPU shows results of approximately at the same level, and we conclude that there are no noticeable changes in vertex sampling tests when moving from Cypress to Cayman.

3DMark Vantage: Feature tests

Synthetic tests from 3DMark Vantage can show us something we previously missed. The Feature tests of this test package support DirectX 10 and are interesting because they differ from ours. When analyzing the results of the new video cards in this package, we will be able to draw some new and useful conclusions that eluded us in the RightMark family tests. Unfortunately, the company's even newer test package - 3DMark11 - does not contain specialized synthetic tests and is not at all interesting to us in this case.

Feature Test 1: Texture Fill

The first test is a texture fetch speed test. This involves filling a rectangle with values ​​read from a small texture using multiple texture coordinates that change every frame.

Although Futuremark's texture test also does not show the theoretically possible level of texture fetch speed, the efficiency of the new video cards of the Radeon HD 6900 family is still slightly higher than in ours. And Nvidia solutions also use existing texture units more efficiently. Therefore, in this texture test we get a slightly different ratio of results compared to ours.

AMD's new family of video cards show results that fully comply with theoretical parameters. The HD 6950 is slightly faster than the HD 5870, and the HD 6970 is the clear winner in the test. You can clearly see that the Cayman's textural performance has increased significantly compared to the Cypress. But the HD 6870 based on the Barts chip shows the worst result, similar to the figures of the top Nvidia video card. Well, the GTX 570 is inferior to everyone in texturing, as in our test.

Feature Test 2: Color Fill

This is a fill rate test. 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.

The situation is completely different in the performance test of ROP blocks. The numbers in this 3DMark Vantage subtest show the performance of ROP units, almost unaffected by the amount of video memory bandwidth. The HD 6970 model shows excellent results, almost catching up with the top-end GTX 580 and ahead of its competitor GTX 570.

In turn, the HD 6950 is also not only ahead of its competitor GTX 570, but also ahead of its predecessor, the HD 5870. We note a slightly higher efficiency of ROP units and a higher fill rate in new models of AMD video cards compared to older chips.

Feature Test 3: Parallax Occlusion Mapping

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 a heavy Strauss algorithm. This is a test of a very complex and heavy pixel shader for a video chip, containing numerous texture samples during ray tracing, dynamic branching and complex lighting calculations according to Strauss.

This test differs from other similar ones in that the results in it depend not exclusively on the speed of mathematical calculations or the efficiency of branch execution or the speed of texture fetches, but on a little bit of everything. And to achieve high speed, the balance of GPU blocks is important. Affects the speed and efficiency of branching in shaders.

The comparison chart between AMD graphics cards is quite similar to what we saw in the texture performance test from 3DMark Vantage. But Nvidia boards in this case received a slight increase in performance, which suggests that not only texture performance affects the test results.

New AMD models have once again seriously asserted themselves, overtaking their predecessor in the form of the HD 5870. But the HD 6870 from a different price sector showed a noticeably weaker result, becoming an outsider in this test (which is completely mitigated by its low price). As for comparing the Cayman with competing Nvidia solutions, both new video cards of the HD 6900 family are ahead of even the top model in the GeForce GTX 500 line.

Feature Test 4: GPU Cloth

The 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 depends on many parameters, but the main ones are geometry processing performance and geometry shader execution efficiency. It is logical that video cards produced by Nvidia feel like fish in water in this application, and are significantly ahead of their competitors, including the top models presented today.

This is one of the few tests without tessellation, in which the advantage of the recently introduced video cards of the new Radeon HD 6800 series and today's heroes HD 6900 is visible. The rendering speed of all these models in this test is higher than that of the top model of the previous line. This is because both Barts and Cayman have increased the speed of geometry processing and geometry shader execution. And although even the HD 6970 continues to seriously lag behind the GTX 570, the new model has still significantly improved AMD's position in this test.

Feature Test 5: GPU Particles

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 most of all loads shader units with vertex calculations; stream out is also tested.

The results of the next test from the 3DMark Vantage package are similar to those we saw in the previous diagram, but in it the speed of geometry processing is even more important. That is why the previous generation in the form of the Radeon HD 5870 card lagged behind both GeForce models, which are the undisputed leaders in comparison, and all new models of AMD video cards, the HD 6900 and HD 6800 families. And all boards based on Cayman and Barts showed more better results than the only solution on Cypress, second only to strong competitors.

It appears that in the synthetic fabric and particle simulation tests from the 3DMark Vantage benchmark suite, which make extensive use of geometry shaders, there is again no significant impact of parallelized geometry processing on the Cayman, as Barts showed a similar result. Therefore, both solutions of the HD 6900 line continue to lag behind competing video cards of their opponents, which have excellent geometry processing speed - up to two times higher. We still expected a little more progress from AMD's top solution, based on a new architecture with two geometry processing units.

Feature Test 6: Perlin Noise

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 it uses a lot of mathematical calculations.

In a purely mathematical test from the Futuremark package, showing the peak performance of video chips in extreme tasks, we saw an even more interesting picture than in similar tests from our test package. The performance of the solutions shown in the diagram only approximately corresponds to what should be obtained according to theory and is somewhat at odds with what we saw earlier in mathematical tests from the RightMark 2.0 package.

Even from the theoretical characteristics of the new HD 6970 and HD 6950 models, it was clear that they did not increase the peak performance of mathematical calculations compared to the HD 5870. But still, we did not expect a clear lag. Yes, both boards outperformed their competitors from Nvidia by a huge margin, but we are used to this, because GeForce video cards do not show very good results in such cases; simple and intensive math is performed significantly faster on Radeon.

What is unexpected is that the new older model lost 7% to the previous top model, although theoretically it should lose no more than 1%. Here again one can begin to guess about what caused this loss to his predecessor. Either this is due to the lack of driver optimization for new solutions, or the lower efficiency of the VLIW4 architecture in such tests, or an overly smart power management system on new models, which “killed” the clock frequency and performance of solutions when the set power consumption threshold is reached.

Direct3D 11: Compute Shaders

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

First, let's look at tests 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 , using pixel and compute shaders.

Perhaps this example for computational shaders is not the best, but there are very few of them yet. All video cards show similar results in this test, but the top model Geforce GTX 580 still wins. The cards announced today on the new Cayman chip are only slightly inferior to it, and only when using a pixel shader. The direct competitor of AMD's new solutions, the GTX 570 video card, lags behind them in both modes: both using pixel and using compute shaders.

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

But here are more interesting results, for AMD solutions somewhat similar to the numbers from the 3DMark Vantage mathematical test. Despite the large theoretical advantage in peak numbers, the fastest Radeon HD 5870 graphics card is only slightly ahead of Nvidia's best solution. And both new models of the HD 6900 family show results close to those of their direct competitor - Geforce GTX 570.

But today we are more interested in the difference between the results of solutions on Cayman and Cypress, and here we again see how the old video card wins, and with what an advantage! 17% between HD 5870 and HD 6970 in favor of the former - once again mathematical tests reveal the difference between beautiful theory and brutal practice. Well, where are the applications in which the new GPU can show its strength? Perhaps in tessellation tests everything will finally fall into place.

Direct3D 11: Tessellation Performance

Compute shaders are very important, but the main innovation in Direct3D 11 is still hardware tessellation. We looked at it in great detail in our theoretical article about the Nvidia GF100. Tessellation has long been used in DX11 games, such as STALKER: Call of Pripyat, DiRT 2, Aliens vs Predator, Metro 2033, Civilization V and others. Some of them use tessellation for character models (all of the FPS games listed), others - to simulate a realistic water surface (DiRT 2) or landscape (Civilization V).

There are several different schemes for partitioning graphic primitives (tessellation). For example, phong tessellation, PN triangles, Catmull-Clark subdivision. The PN Triangles tessellation 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 shows 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:

Let's look at pixel-by-pixel techniques first. Parallax occlusion mapping (middle bars in the diagram) on video cards from both manufacturers is much less efficient than tessellation (lower bars), and moderate tessellation does not give a big drop in performance - compare the upper and lower columns. That is, high-quality imitation of geometry using pixel calculations provides even lower performance than tessellated geometry with displacement mapping.

As for the performance of video cards relative to each other, the most important conclusion here is that AMD video cards are slightly faster than Nvidia cards in the easiest mode, but slower in complex pixel-by-pixel calculations (remember the parallax mapping tests earlier). And before the release of Cayman boards, Geforce cards were slightly faster than AMD solutions even with tessellation enabled.

But now the difference in geometry processing speed between the HD 6900 and HD 5870 is clearly visible - the new Cayman-based boards turned out to be noticeably faster than the Cypress in the tessellation subtest. In this test with a small triangle split ratio, the HD 6970 even outperformed its competitor, the GTX 570, by a good margin.

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.

Only in this example do we see for the first time a truly complete comparison of the geometric power of AMD and Nvidia solutions. Both the Fermi graphics architecture and the new AMD GPU under the name Cayman stand out very much. Putting aside the fact that this is a purely synthetic test and such extreme partitioning ratios would not be used in today's games, we are now interested in the potential. After all, synthetics are needed in order to evaluate the prospects and differences between different solutions.

It is immediately clear that it is impossible to compete with Nvidia Geforce video cards on the GF110 chip; in extreme tessellation tasks they are many times faster than even the updated AMD architecture. But this is an architecture specifically designed from the outset taking into account the capabilities of the new API. What about the Cayman? Compared to Cypress, everything is very good!

New AMD models show an impressive increase in speed in medium load modes, and the difference compared to the HD 5870 reaches more than two times. However, we do not always see such an increase, and most often it falls within one and a half times. At least, we definitely didn’t see the promised threefold difference. That is, although the Cayman has narrowed the gap from its competitor in geometry processing tasks, it is still very far from the parallel operation of 16 tessellation units in the GF110.

On the other hand, the greatest difference between solutions from different companies is achieved in conditions of extreme tessellation, which do not exist and are not yet expected in real games. Therefore, most likely, Cayman will significantly strengthen AMD's position in existing gaming benchmarks using tessellation. Especially if the partitioning factor is not too large, as in the 3DMark11 tests.

Let's look at 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. It looks just great, here's what's missing in current games:

Island is not a purely synthetic test for measuring geometry performance; it contains complex pixel and compute shaders, so the difference in performance may be less than in the previous case, but this load is closer to real games that use all GPU blocks at once.

We tested the program at four different tessellation ratios, this setting is called Dynamic Tessellation LOD. If at the lowest partitioning factor AMD video cards are ahead, then as the work becomes more complicated, boards based on GF110 immediately jump far ahead. As the partitioning factor and scene complexity increase, the performance of all Radeons drops very significantly, in contrast to the speed of competing solutions.

Moreover, this time the HD 5870 for some reason is even ahead of both models of the new family. That is, there is a difference opposite to theory in a problem with complex geometry. And there can be only one explanation for this - the lack of driver optimization for the new architecture, because in previous tests we saw its clear advantage over the Radeon HD 5870, based on the Cypress chip. Well, in this test we are so far forced to admit the defeat of the Cayman - at the maximum LOD coefficient, the difference between the speed of the GeForce and the Radeon reached 4-6 times!

Conclusions on synthetic tests

Based on the results of synthetic tests of video cards from the new Radeon HD 6900 family, based on the Cayman graphics processor, as well as the results of other video card models produced by both manufacturers of discrete video chips, we can conclude that the new products are a good replacement for the Radeon HD 5800 line, although not too much different from it in terms of performance, at least in synthetic tests.

The Cayman GPU is based on a new architecture and differs from previous chips in hardware, although the number of some execution units in it has not increased. But the new GPU features architectural improvements aimed at increasing the efficiency of GPU computing (we don’t even have such tests) and, more importantly, mitigating the important gap from the competitor in the form of geometry processing performance. Many of the synthetic tests show that the speed of tessellation and execution of geometry shaders has noticeably increased, although not always several times as promised.

Thanks to architectural changes and their frequency characteristics, the results of video cards of the new series in many synthetic tests are competitive for their price sector, especially compared to the direct competitor Geforce GTX 570. This is even more clearly visible in computing tests from the RightMark and Vantage packages. And in other applications, the solutions of the HD 6900 family showed good speed, most often second only to the top Nvidia video card.

Unfortunately, there were some not very pleasant surprises. Despite the greater complexity and chip area compared to the Cypress, the HD 6900 models performed lower than the HD 5870 in some math tests, which is not easy to explain, and we are not yet sure of the reasons for this lag. Perhaps the lack of driver optimization is to blame, or perhaps the efficiency of the new VLIW4 architecture was lower in our tests. It is also likely that the power management system on the new models lowered the clock speeds when reaching maximum power consumption in demanding synthetic tests, not allowing them to show the performance expected based on the number of blocks and their clock speed.

Surely many expected that the Radeon HD 6970 would be able to compete on equal terms with the GTX 580 in all tests, but this did not happen, although the results were shown to be very good and quite consistent with the recommended prices for the models announced today. We assume that the results of the Radeon HD 6970 and HD 6950 in synthetic tests will be confirmed by the corresponding figures in the “game” part of our material. In games, the older HD 6970 should perform approximately at the level of the GTX 570, in some tests a little slower, and in others - faster, and the HD 6950, although it will be slower than this Nvidia model, is also priced lower. So let's quickly move on to exploring speed in games!

Introduction While Nvidia's release of its new mid-range GF114-based solution was a shake-up in the $250 performance graphics card niche, AMD's graphics division found itself in an awkward position with the introduction of the GeForce GTX 560 Ti. The Radeon HD 6950, capable of competing with the new “green” product, cost $299, and the Radeon HD 6870, at a comparable price, was clearly inferior to the GeForce GTX 560 Ti. To the company’s credit, it did not lose its head and responded to Nvidia’s attack with two smart maneuvers: reducing the price of the Radeon HD 6870 from $239 to $219 and releasing a cheaper version of the Radeon HD 6950 costing $259. Today’s review will be devoted to the latter, so let’s go a little deeper into the details.

The amount of video memory on board gaming graphics cards has grown linearly as consumer 3D has progressed. For quite a long time, the question “do we need more megabytes?” was one of the most regularly asked questions. At first it was about the need for 128 MB compared to the mass 64 MB, then serial cards began to be equipped with 256 MB of video memory, but they began to talk about 512 MB, and finally, most game cards began to be equipped with 1 GB of local memory, after which the process practically stopped. And no wonder - 1 GB of video memory, when used correctly, is enough in 99% of cases and usage scenarios, especially taking into account the fact that an increasing number of game projects are multi-platform, and therefore have less serious requirements for this parameter. Of course, older Nvidia GeForce GTX 500 models carry more on board, however, this is due to the configuration of their memory controllers, which form 320- or 384-bit external access buses.

What is less clear is the reason why AMD, when releasing the new Radeon HD 6900 family, equipped both of its representatives, the Radeon HD 6970 and Radeon HD 6950, with 2 GB of video memory, which only increased the cost of the final products and made the younger model much less competitive than it could have been be. Fortunately, the company quickly realized its mistake, and the new $259 Radeon HD 6950 model received 1 GB of video memory, while retaining all the other characteristics of its more expensive brother.

We got our hands on one of these cards, released by PowerColor, so now all that remains is to make sure that 2 GB of video memory on board a modern high-end gaming video card is overkill. PowerColor HD6950 1GB will help us in finally clarifying this issue.

PowerColor HD6950 1GB: design and technical specifications

The manufacturer clearly has a weakness for vertical boxes, since the HD6950 1GB comes in almost the same packaging as the previously described HD6870 PCS+. Only the design on the front side has changed.

The motifs of the drawing are clearly inspired by the Warhammer 40K universe, which not everyone will like, but, as we have repeatedly repeated, this is purely a matter of taste. Let's just say that in this case there is clearly no need to talk about any originality. There is also little useful information on the packaging - only the model of the video adapter is indicated, as well as the type and amount of video memory installed on board.

The inside of the box is an ordinary cardboard pallet, so there is no talk of serious protective properties. In addition to the video adapter in an antistatic bag, you can find the following set of related accessories:

DVI → D-Sub adapter
Mini DisplayPort → DisplayPort adapter
Installation Guide
CD with drivers and utilities

The equipment is no less spartan than in the case of the HD6870 PCS+; There is not even a CrossFire connecting bridge. However, this only has a positive effect on the availability of the product, which for many fans of modern games is a very important factor, along with performance.

Paradoxically, the PowerColor HD6950 1GB does not use the reference PCB design developed by AMD for the Radeon HD 6950, but a slightly redesigned design of the Radeon HD 6870. This approach on the part of PowerColor can be called, at least, original. We saw almost the same design in the HD6870 PCS+ design:

Of course, there are differences, but the cooling system makes it difficult to see them, so we did not fail to dismantle it by unscrewing the four spring-loaded screws:

It turned out that there are, in fact, very few differences: a slightly different ground layout in the rear part of the board and changes in the upper left corner caused by the need to install a second CrossFire connector and a BIOS switch. There are others, but it is obvious that the PowerColor developers tried to make the most of the developments already at their disposal.

The four-phase power regulator for the graphics core uses a CHiL Semiconductor CHL8214 controller, but the memory power subsystem contains an unknown controller manufactured by uPI Semiconductor with the strange marking uP1509P. The general layout of the power part has not changed; it is still built according to the “4+2” scheme, and external power is connected to the board via two six-pin PCIe 1.0 connectors.

The memory chips used are GDDR5 H5GQ1H24AFR chips with a capacity of 1 Gbit (32Mx32), well known to us and our readers, produced by Hynix Semiconductor. The T2C suffix indicates a nominal frequency of 1250 (5000) MHz, at which the memory operates, providing a throughput of the entire subsystem of 160 GB/sec. One can hardly expect serious achievements from it in terms of overclocking, but the PowerColor HD6950 1GB will not suffer from a lack of bandwidth anyway.

The GPU was produced in week 49 of last year, 2010. Of course, we are talking about a truncated version of the Cayman, in which only 1408 ALUs out of 1536 physically available on the GPU chip are active. Thus, 352 universal VLIW4 processors out of 384 are active. The texture processing subsystem is truncated from 96 to 88 texture processors, but the raster subsystem is not affected: the configuration of memory controllers is strictly tied to it, so all 32 RBE units are active.

The core clock frequency in power-saving mode is 250 MHz, and in full load mode - 800 MHz, which fully complies with the official AMD specifications for the Radeon HD 6950. The core supply voltage can be one of three values: 0.898 V, 1.0 V or 1.063 V , depending on the operating mode. There is a BIOS switch in the upper left part of the board, with which you can switch between protected factory firmware and free firmware that can be flashed by the user.

The PowerColor HD6950 1GB mounting strip demonstrates the standard configuration of switching connectors. It carries a pair of DVI-I ports, an HDMI port and a pair of Mini DisplayPort connectors. Of course, all five connectors can be used simultaneously, and with the use of a switch that supports the DP 1.2 standard, the number of simultaneously operating displays can be increased to six. A pair of CrossFire connectors allows you to build a powerful CrossFireX system consisting of four cards. With the advent of the Radeon HD 6990 on the market, this opportunity is not very relevant, but it is still present.

As noted in the review of the PowerColor HD6870 PCS+, the cooling system installed on this card turned out to be extremely unsuccessful or defective. In the case of the HD6950 1GB we see a similar design:

The radiator area is significantly larger, but there are still two heat pipes. However, the radiator is already blown by two 92 mm fans. If it were not for the results of previous tests of a similar design, one could say that the PowerColor HD6950 1GB cooling system looks promising. But given the previously obtained data, we consider it reasonable to exercise additional caution in our estimates. However, the next chapter will put everything in its place.

Power consumption, thermal conditions, noise and overclocking

The described Radeon HD 6950 model, equipped with 1 GB of local video memory, is encountered in our practice for the first time, so we decided to spend a little time figuring out the issue of the energy consumption level of this solution. To do this, it was installed on a measuring stand with the following configuration:

Processor Intel Core 2 Quad Q6600 (3 GHz, 1333 MHz FSB x 9, LGA775)
DFI LANParty UT ICFX3200-T2R/G motherboard (ATI CrossFire Xpress 3200)
Memory PC2-1066 (2x2 GB, 1066 MHz)
Enermax Liberty ELT620AWT Power Supply (Rated Power 620W)
Microsoft Windows 7 Ultimate 64-bit
CyberLink PowerDVD 9 Ultra/“Serenity” BD (1080p VC-1, 20 Mbps)
Crysis Warhead
OCCT Perestroika 3.1.0

This platform is equipped with a special measuring module, described in the review “ Computer power consumption: so how many watts do you need?" Its use allows you to obtain the most complete data on the electrical characteristics of modern graphics cards in various modes. As usual, the following tests were used to load the video adapter in various modes:

CyberLink PowerDVD 9: FullScreen, hardware acceleration enabled
Crysis Warhead: 1600x1200, FSAA 4x, DirectX 10/Enthusiast, frost card
OCCT Perestroika GPU: 1600x1200, FullScreen, Shader Complexity 8

For each mode, with the exception of the ultimate load simulation in OCCT, measurements were taken for 60 seconds; To avoid card failure due to power supply overload, for the OCCT: GPU test the testing time was limited to 10 seconds. As a result of the measurements taken, the following results were obtained:

Let's be brief: measurements of the power consumption level of the PowerColor HD6950 1GB did not reveal anything new. In 2D mode and HD video decoding, the card consumes a little more than the reference Radeon HD 6950 2GB, and in 3D mode, on the contrary, a little less. 161 W is the level of the GeForce GTX 560 Ti, so the product described is quite competitive in its class in terms of power consumption.

Note that when decoding high-definition video, the figures show a peak level of power consumption, which is clearly visible from the above diagram. The average in this mode is approximately 40 W. The second power connector, closest to the edge of the printed circuit board, designated in the table as “12V 6-pin,” is loaded noticeably more than the first, but there is no serious excess of the 75 Watt recommended for this type of connector even in OCCT.

Let's return to a more pressing issue, namely, thermal and noise characteristics. At room temperature within 25 degrees Celsius, we were able to obtain the following results:

The indicators, at first glance, are not outstanding, and are not even very different from the indicators of the PowerColor HD6870 PCS+, however, firstly, the HD6950 1GB has a significantly higher level of heat dissipation, and secondly, compared to the indicators of the reference model Radeon HD 6950 2GB, the system cooling PowerColor HD6950 1GB looks very, very good! But, as we remember, the PowerColor HD6870 PCS+ had a fairly high cooling efficiency that was accompanied by a monstrous noise level. Will this happen again with the PowerColor HD6950 1GB?

This, oddly enough, depends on the position of the BIOS switch. If protected factory firmware is used, the noise level is within reasonable limits; in any case, with a background noise level in the room of 38 dBA, the performance of the PowerColor HD6950 1GB is no worse than that of the reference Radeon HD 6950 2GB. At least this is true for 3D mode. But switching to the “enthusiast” position instantly leads to a significant increase in fan speed and a corresponding increase in noise level. Apparently, this is a kind of protection against user experiments with overclocking and increasing the GPU supply voltage. Of course, it won’t save you from real extreme sports enthusiasts, however, the idea of ​​reinsurance in general does not seem superfluous; fortunately, real overclockers, as a rule, do not care about acoustic comfort. Of course, all test results, including temperature ones, were obtained in “b1” mode, which means secure firmware.

The results of overclocking the PowerColor HD6950 1GB turned out to be quite modest. After a number of attempts, we managed to achieve stable operation of the card at the following frequencies:

As you can see, only 860 MHz graphics core frequency can hardly be called serious overclocking. The memory was somewhat more pleasing, reaching an impressive 5400 MHz equivalent GDDR5 frequency. As a result, after some hesitation, it was decided to test the PowerColor HD6950 1GB not only at factory frequencies, but also in overclocking mode.

Test platform configuration and testing methodology

Testing of the PowerColor HD6950 1GB in real gaming applications was carried out on a universal test platform with the following configuration:

Processor Intel Core i7-975 Extreme Edition (3.33 GHz, 6.4 GT/s QPI)
Cooler Scythe SCKTN-3000 “Katana 3”
Gigabyte GA-EX58-Extreme motherboard (Intel X58)
Memory Corsair XMS3-12800C9 (3x2 GB, 1333 MHz, 9-9-9-24, 2T)
Hard drive Samsung Spinpoint F3 (1 TB/32 MB, SATA II)
Power supply Ultra X4 850W Modular (Rated power 850 W)
Monitor Dell 3007WFP (30”, maximum resolution 2560x1600@60 Hz)
Microsoft Windows 7 Ultimate 64-bit

The following versions of ATI Catalyst and Nvidia GeForce drivers were used:

ATI Catalyst 11.4 “Mjölnir I” preview for ATI Radeon HD
Nvidia GeForce 266.66 WHQL for Nvidia GeForce GTX 560 Ti
Nvidia GeForce 266.58 WHQL for Nvidia GeForce GTX 570

The drivers themselves were configured as follows:

ATI Catalyst:

Anti-Aliasing: Use application settings/4x/Standard Filter
Morphological filtering: Off
Tesselation: Use application settings
Texture Filtering Quality: High Quality
Enable Surface Format Optimization: Off
Wait for vertical refresh: Always Off
Anti-Aliasing Mode: Adaptive Multi-sample AA

Nvidia GeForce:

Texture filtering - Quality: High quality
Vertical sync: Force off
Antialiasing - Transparency: Multisampling
CUDA - GPUs: All
Set PhysX configuration: Auto-select
Ambient Occlusion: Off
Other settings: default

The test package included the following games and applications:

3D first-person shooters:

Aliens vs. Predator (1.0.0.0, Benchmark)
Battlefield: Bad Company 2 (1.0.5, Fraps)
Call of Duty: Black Ops (1.04, Fraps)
Crysis Warhead (1.1.1.711, Benchmark)
Metro 2033 (Ranger Pack, 1.02, Benchmark)
S.T.A.L.K.E.R.: Call of Pripyat (1.6.02, Fraps)

3D shooters with third person view:

Just Cause 2 (1.0.0.1, Benchmark/Fraps)
Lost Planet 2 (1.1, Benchmark/Fraps)

RPG:

Dragon Age II (1.01, Fraps)
Mass Effect 2 (1.01, Fraps)

Simulators:

F1 2010 (1.01, Fraps)

Strategy games:

StarCraft II: Wings of Liberty (1.3, Fraps)
Total War: Shogun 2 (1.1, Fraps)

Semi-synthetic and synthetic tests:

Futuremark 3DMark Vantage (1.1)
Futuremark 3DMark 11 (1.0.1)
Unigine Heaven Benchmark (2.5)
Tom Clancy's H.A.W.X. 2 Benchmark (1.04, Benchmark/Fraps)

Each of the games included in the test software was configured to provide the highest possible level of detail. Applications that support tessellation have enabled this feature.

The fundamental refusal to manually modify any configuration files means that only the tools available in the game itself to any uninitiated user were used for configuration. Testing was carried out in resolutions of 1600x900, 1920x1080 and 2560x1600. Except where otherwise noted, standard 16x anisotropic filtering was complemented by 4x MSAA anti-aliasing. Activation of anti-aliasing was carried out either by the game itself, or, in their absence, it was forced using the appropriate settings of the ATI Catalyst and Nvidia GeForce drivers.

In addition to the PowerColor HD6950 1GB, the following graphics cards took part in testing:

ATI Radeon HD 6950 2GB
ATI Radeon HD 6870
Nvidia GeForce GTX 570
Nvidia GeForce GTX 560 Ti

To obtain performance data, we used the testing tools built into the game with the obligatory use of original test videos, as well as, if possible, recording data on minimum performance. In some cases, the Fraps 3.3.2 test utility was additionally used to obtain information about minimum performance. In the absence of the above-mentioned tools, the same utility was used in manual mode with a three-fold test pass, fixing the minimum values ​​and then averaging the final result.

Game tests: Aliens vs. Predator

As we expected, the Radeon HD 6950 did not suffer at all from cutting the amount of local video memory in half, except for the dropped price, which, of course, is difficult to attribute to losses, especially from the buyer’s point of view. The PowerColor card feels very confident both in normal mode and with additional overclocking, and, in the latter case, the HD6950 1GB can compete with the more expensive GeForce GTX 570. Only the 2560x1600 mode remains unavailable, requiring higher-class solutions.

Game tests: Battlefield: Bad Company 2

There is no difference between the two Radeon HD 6950 models, which is good news. It seems that today the desktop 3D graphics industry has reached a point where increasing the amount of video memory will not make sense for quite some time. In this game, the GeForce GTX 560 Ti is in the lead, but not enough to definitely prefer this card over the Radeon HD 6950, especially since overclocking allows you to squeeze a little more performance out of the latter.

Game Tests: Call of Duty: Black Ops

The game is absolutely undemanding for fairly modern graphics cards, however, it is clear that among all test participants, the GeForce GTX 560 Ti demonstrates the worst minimum performance, while both versions of the Radeon HD 6950 perform excellently even in Ultra HD modes. Accordingly, there is no need to resort to overclocking.

Game tests: Crysis Warhead

So far, this is the only test where we were able to detect at least some difference in the performance of the two Radeon HD 6950 models, albeit very insignificant and only at a resolution of 2560x1600, where the comparison loses its meaning due to low performance. The PowerColor product looks clearly more profitable than the GeForce GTX 560 Ti, since, in contrast, it provides quite acceptable performance in Full HD mode, both with and without overclocking. Approximately the same performance can be obtained on the GeForce GTX 570, however, it will cost you much more.

Game tests: Metro 2033

It is interesting that overclocking the Radeon HD 6950 does not have any noticeable effect in this game, but in terms of overall performance this video adapter is second only to the GeForce GTX 570. The only resolution in which the difference is noticeable is 1600x900, so there is no reason to overpay. However, we should not forget about serious fluctuations in the minimum speed when tessellation is enabled.

Game tests: S.T.A.L.K.E.R.: Call of Pripyat

The game is tested with tessellation enabled.

The effect of overclocking is small and can be ignored, especially since the performance of the Radeon HD 6950 is already sufficient for a full game at a resolution of 1920x1080. Along with the GeForce GTX 560 Ti, the version of the Radeon HD 6950 with 1 GB of video memory is deservedly one of the best mainstream gaming cards of a fairly high class.

Game tests: Just Cause 2

As before, there is no point in buying a Radeon HD 6950 2GB. The model, equipped with half the amount of local video memory, works no worse across the entire resolution range. There was also no particular sense in overclocking, although at a resolution of 1600x900 its use allowed the hero of our review to catch up with the GeForce GTX 570.

Game tests: Lost Planet 2

It has already been proven that Lost Planet 2 openly prefers the Nvidia Fermi architecture, so AMD solutions have practically nothing to do in this game. Even a fairly powerful Radeon HD 6950 can only be used at a resolution of 1600x900, and to correct the situation, a clearly more significant overclock will be required than what we were able to achieve in the case of the PowerColor HD6950 1GB.

Game Tests: Dragon Age II

The new RPG released by Bioware was not without reason used in advertising the Radeon HD 6990. Crysis Warhead now has a truly serious opponent in terms of appetite for graphics subsystem resources. Interestingly, the situation here looks exactly the opposite in comparison with Lost Planet 2 - Nvidia solutions are clearly in the fold, and cards based on AMD Cayman rule the roost. At maximum detail settings, the hero of our review allows you to play quite comfortably at a resolution of 1600x900, which, given the remarkable appetites of this game, is in itself a significant achievement, especially if you look at the deplorable performance of the GeForce GTX 570. As in other games, with the exception of the same Crysis Warhead, the difference in video memory capacity of the two Radeon HD 6950 models does not have any effect.

Game Tests: Mass Effect 2

In this test, full-screen anti-aliasing is boosted using the technique described in the Contemporary Graphics Accelerators in Mass Effect 2 review.

In this case, overclocking the PowerColor HD6950 1GB allows it to catch up with the GeForce GTX 560 Ti at a resolution of 1600x900 and surpass it at higher resolutions, but even without this measure, the performance of the product under review is quite high in all modes, including Ultra HD. Along with the GeForce GTX 560 Ti, the video adapter described is, without a doubt, an excellent choice for Mass Effect 2.

Game tests: F1 2010

The PowerColor product performs no worse in the Formula 1 racing simulator; in any case, it provides a noticeably higher level of average and minimum performance than the GeForce GTX 560 Ti, only slightly inferior to the noticeably more expensive GeForce GTX 570.

Game Tests: StarCraft II: Wings of Liberty

The review is nearing the end, but so far StarCraft II can be called the only game where overclocking the PowerColor HD6950 1GB brings at least some benefit, namely, it slightly increases performance at a resolution of 2560x1600. Since the increase is small, the reality of this benefit can be debated, but it should be noted that the capabilities of the product under study provide a fairly high level of comfort even at this resolution. Overall, the Radeon HD 6950 1GB looks only slightly worse than the GeForce GTX 570.

Game tests: Total War: Shogun 2

The features of this game's engine force you to choose between using FSAA and a full set of special effects. We opted for the latter.

For some reason, the minimum performance of the GeForce GTX 560 Ti is below all criticism with average performance comparable to that of the Radeon HD 6950 1GB, which determines the preferred choice for a potential buyer who does not want to shell out money right away for a GeForce GTX 570. It should be noted right away that neither one of the test participants was unable to achieve acceptable performance in the 2560x1600 resolution.

Semi-synthetic and synthetic tests: Futuremark 3DMark Vantage

To minimize the impact of the CPU, testing in 3DMark Vantage uses the “Extreme” profile, which uses a resolution of 1920x1200, FSAA 4x and anisotropic filtering. To complete the performance picture, individual test results are captured across the entire resolution range.

The results are quite logical: the performance of the two Radeon HD 6950 models, if at all different, is minimal, and the overclocking of the PowerColor card is modest enough to bring significant dividends. As a result, the GeForce GTX 570 is left to rest on its laurels, while the unoverclocked GeForce GTX 560 Ti and Radeon HD 6870 trail behind.

All of the above is true for individual tests, except that in the second test the lag between the GeForce GTX 560 Ti and the Radeon HD 6950 is not as significant as in the first, and at a resolution of 2560x1600 we can already talk about parity between these two cards.

Semi-synthetic and synthetic tests: Futuremark 3DMark 11

In the new Futuremark benchmark suite we also use the "Extreme" profile, however, unlike 3DMark Vantage, in 3DMark 11 it uses a resolution of 1920x1080.

The performance of two Radeon HD 6950, one of which has 1 GB of video memory on board, and the other twice as much, is absolutely the same. A difference of ten to one and a half points can easily be attributed to random fluctuations. However, unlike 3DMark Vantage, in 3DMark 11, using overclocking, you can raise the result of the card under study almost to the level of the GeForce GTX 570.

Semi-synthetic and synthetic tests: Tom Clancy's H.A.W.X. 2 Preview Benchmark

This test actively uses tessellation to render the earth's surface. The number of polygons in a frame can reach one and a half million.

Unlike the Radeon HD 6800, the Radeon HD 6900 series flies clearly higher, but still not as high as Nvidia Fermi solutions, which have much more impressive geometry processing and tessellation capabilities. However, the indicators are quite optimistic even at a resolution of 2560x1600. Overclocking adds quite a bit to the existing result.

Semi-synthetic and synthetic tests: Unigine Heaven benchmark

The test uses the “normal” tessellation mode.

Despite the relatively modest tessellation engine compared to solutions based on Nvidia Fermi, the Radeon HD 6950 demonstrates very good results, only slightly inferior to the results of the GeForce GTX 570. Moreover, in resolutions from 1920x1080 and higher in the new version of Unigine Heaven GeForce GTX 560 Ti sharply loses in minimum performance, which plays into the hands of this AMD solution.

PowerColor HD6950 1GB: advantages and disadvantages

Advantages:

High level of performance in its class
Wide range of FSAA modes
Industry-leading anisotropic filtering performance
Supports six monitor output
Full hardware support for HD video decoding, including DivX and 3D
High-quality post-processing and scaling of HD video
Integrated sound engine with support for HD audio formats
Support audio output via HDMI
HDMI 1.4a support
DisplayPort 1.2 support
High cooling efficiency
Low noise level

Flaws:

Modest overclocking potential
Fewer selection of GPGPU applications than competing solutions

Conclusion

As expected, reducing the amount of video memory on the Radeon HD 6950 did not have any visible effect. Only at a resolution of 2560x1600 in Crysis Warhead and StarCraft II an extremely insignificant drop in performance was found, not exceeding 9 and 6%, respectively. In other cases, if a lag was observed, it did not even reach 5%. In other words, we can talk about complete parity between the Radeon HD 6950 1GB and Radeon HD 6950 2GB.

In this regard, the question arises - what is the scope of application of the second, more expensive model? We cannot answer this question unequivocally, but one thing is clear - this is clearly not the gaming sphere, since in 2011, in the vast majority of games, 1 GB of local video memory is quite enough, and this figure has long been established as a kind of standard. Perhaps in some applications that perform massive calculations on the GPU, 2 GB of local memory may be in demand, but for such purposes there are specialized solutions equipped with even larger amounts of memory and, accordingly, better suited for such tasks.

It's obvious that the Radeon HD 6950 2GB won't survive in the gaming graphics card market: at $299, it's topped by the faster GeForce GTX 570 and bottomed by the less expensive GeForce GTX 560 Ti. But the Radeon HD 6950 1GB fits very well into the niche between the two aforementioned Nvidia solutions, being significantly cheaper than the first and significantly faster than the second. Performance summary charts can serve as confirmation:

First of all, the Radeon HD 6950 1GB can boast that it is 8-30% ahead of the GeForce GTX 560 Ti on average, depending on the resolution, and in some tests the superiority can be twofold. The latter, however, does not speak in favor of AMD, but rather that some game prefers a certain graphics architecture. The gap from the GeForce GTX 570 is small; on average, it is 3-10%, depending on the resolution, but in some tests it can reach 40%, so the preferred choice of a particular player, as before, depends on his preferred set of games. As for overclocking, as expected, such a modest increase in GPU and memory clock speeds brought a corresponding increase in performance - from 2 to 14%, depending on the game and resolution.

Regarding the specific Radeon HD 6950 1GB model discussed in this review, PowerColor HD6950 1GB, we can safely say that it is worth its price in full. This video card does not demonstrate extraordinary talent in the field of overclocking and cannot boast of a rich package. All it does is provide the above-described level of performance in modern games, while demonstrating good thermal and acoustic characteristics. In other words, the PowerColor HD6950 1GB is a good “workhorse” that does its job well. It does not have outstanding advantages, but it is also devoid of visible shortcomings, which means that we can fully recommend it to anyone who is looking for a productive gaming card at a reasonable cost.

Other materials on this topic

Review of the GeForce GTX 550 Ti accelerator and MSI N550GTX-Ti Cyclone OS video card
Reference tables for ATI, AMD and NVIDIA video solutions
MSI R6870 Hawk vs PowerColor HD6870 PCS+: review of two Radeon HD 6870 models

About a year has passed since the announcement of older video cards based on the Cypress graphics core. And as if according to the schedule of Japanese high-speed trains, the successor to successful architecture is presented to the world. Over the course of a year, AMD gathered its strength and made a small revolution, changing the very heart of its GPUs: five former scalar processors were optimized into a dual pair. As well as the film of the same name, Nvidia’s countermeasures, the successful GF104 and the ultra-fast castling in the upper segment - GF100 -> GF110 were taken into account.

advertising

After the launch of the Barts series, there is not much space left in the ranks of cards, and you have to carefully plan placement among both competitive solutions and your own graphics adapters. As a result, from now on we can consider that the Middle…High segment of AMD video cards has been completely updated. HD 5830/50/70, which left the stage, was replaced by HD 6850/70 and 6950/70.

Characteristics table

Characteristics	HD 5870	HD 6870	HD 6950	HD 6970	GTX 470	GTX 480	GTX 570	GTX 580
Codename	Cypress XT	Barts XP	Cayman Pro	Cayman XT	GF100	GF100	GF110	GF110
Technical process, nm	40	40	40	40	40	40	40	40
Size of core/cores, mm 2	334	255	389	389	~500	~500	~530	~530
Quantity transistors, million pieces	215,4	180	264	264	320	320	330	330
2D core frequency, MHz	157	100	150	150	50 / 100	50 / 100	50 / 100	50 / 100
3D core frequency, MHz	850	900	800	880	607 / 1215	701 / 1402	732 / 1464	772 / 1544
OC core frequency, MHz	1050	950	950	950	800 / 1600	875 / 1750	875 / 1750	850 / 1700
Voltage at core 2D, B	0,95	0,95	0,90	0,90	0,88	0,96	0,91	0,96
Voltage at core 3D, B	1,15	1,21	1,28	1,28	0,99	1,01	1,01	1,06
Number of shaders PC. (PS)	1600	1120	1408	1536	448	480	480	512
Number of blocks rasterization, PC. (ROP)	32	32	32	32	40	48	40	48
Number of textures blocks, pcs. (TMU)	80	56	88	96	56	60	60	64
Maximum fill speed, Gpix/sec	27,2	28,8	25,6	28,2	24,3	33,6	29,3	37,1
Maximum speed texture samples, Gtex/sec	68	50,4	70,4	84,5	32,4	42,1	43,9	49,4
Version pixel/ vertex shaders	5.0 / 5.0	5.0 / 5.0	5.0 / 5.0	5.0 / 5.0	5.0 / 5.0	5.0 / 5.0	5.0 / 5.0	5.0 / 5.0
Memory type	GDDR5	GDDR5	GDDR5	GDDR5	GDDR5	GDDR5	GDDR5	GDDR5
Efficient frequency 2D memory, MHz	1200	300	300	300	67	67	67	67
Efficient frequency 3D memory, MHz	4800	4200	5000	5500	3360	3700	3900	4008
Efficient frequency memory OC, MHz	5200	4800	5800	5800	3600	3900	4000	4100
Voltage at memory 2D, B	1,60	1,61	1,60	1,60	1,54	1,58	1,34	1,36
Voltage at memory 3D, B	1,60	1,63	1,60	1,60	1,55	1,58	1,56	1,62
Memory size, MB	1024 / 2048	1024	2048	2048	1280	1536	1280	1536
memory bus, bit	256	256	256	256	320	384	320	384
Bandwidth memory, GB/sec	153,6	134,4	160	176	133,9	177,4	152	192,4
Consumed 2D power, Watt	27	19	20	20	nd	nd	nd	nd
Consumed 3D power, Watt	188	151	250	250	215	250	219	244
Crossfire/Sli	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Card size LxWxH, mm	282x100x38	248x100x37	270x100x37	270x100x37	270x100x38	270x100x38	270x100x38	270x100x38
Recommended price, $	399	239	299	369	249	499	349	499

Judging by preliminary data, the price war should flare up with renewed vigor. An unprecedented number of graphics cards are located in a narrow price range. As a result, one of the producers (or maybe both) will suffer losses, and the buyer will ultimately benefit.

Screens with OLED, AMOLED and even Super AMOLED matrix “burn out” over time. If the same pixels are illuminated on the screen for a long time, they will dim and this will be clearly visible. Typically, virtual navigation buttons, icons in the top bar and a clock are printed. The display cases of smartphones displayed in stores suffer the most from this problem. They are turned on almost around the clock, stand on stands for weeks or months and always show the same demo content, which remains on the matrix forever.

What is causing this problem?

The crux of the problem lies in a key feature of OLED technology. The matrix consists of LEDs of three colors (blue, red and green), and different types of diodes have their own service life. Blue subpixels are less bright, so to maintain color balance, more current is supplied to them than to red and green subpixels. Because of this, the service life of blue diodes is reduced, over time they shine dimmer, and the color rendition of the screen goes into red and green shades.

Burnout occurs in areas where blue or white color is used heavily. Black color does not use the pixel backlight, so it does not cause burn-in. Burnt-in pixels become dark and visible on the screen. The lighter the image, the better they are visible.

Is there a solution?

Manufacturers have not come up with an adequate solution to this problem; they either ignore it altogether or use crutches, providing for periodic displacement of static elements of the operating system interface by several pixels. Users may not notice this shift, but it prevents subpixels from overheating and slows down their deterioration. The matrices of some Samsung smartphones use PenTile technology: blue diodes are larger and glow quite brightly with less current, which increases their service life.

How to avoid burnout?

Burn-in occurs most quickly on a bright screen, so you shouldn’t unnecessarily turn the brightness slider to maximum.

Do not leave your smartphone on with a static image for long periods of time.

Use a dark, or better yet black, theme for your applications and keyboard.

If your smartphone supports themes, change them from time to time.

Change the wallpaper and icon layout on your home screen occasionally.

Do not use your smartphone as a digital watch. There are applications that allow you to display the time on the screen, and a few hours of work in this mode is enough for the pixels under the numbers to burn out.

Is it possible to fix a screen with burnt-out pixels?

Diodes do not recover, so it is impossible to remove burnout from the screen. Some people advise leaving the screen in the sun for a couple of hours. After such a procedure, the burnt-out pixels may not be visible, but not because they have been restored, but because the remaining subpixels have also darkened. If your smartphone is uncomfortable to use due to marks on the screen, it makes sense to take it to a workshop and ask them to replace the matrix or do it yourself.

We have already appeared in the news a couple of times from a bad side. The first time this happened due to problems with the red color, and the second time due to burnout. The latest incident was reported on Reddit and was also mentioned in The Korea Herald.

What is burnout?

Don't confuse the term "display burn-in" with "fire." Sometimes, when viewing an image for a long time, its outlines are imprinted on the display, as if burned into it, even after the screen is locked or another picture is turned on. This problem exists in almost all displays, but it should be noted that modern technologies cope with this successfully.

Samsung warns its LED TV users not to leave the same image on for two hours or more. However, this does not make it any easier, because the manufacturer is simply playing it safe once again, rather than correcting the problem.

How does display burn-in affect the Galaxy S8?

In the case of the Galaxy S8, the flagship uses the Always On Display feature. This way, some parts of the display remain illuminated even if the screen has been locked. We're talking about the clock and the Home button. Yes, if desired, the Always On Display function can be disabled, but the display still works for several hours in one mode. This is why if a static image is left on, the LEDs will reproduce it. Shouldn't the S8 and S8 Plus be immune to this problem?

This is not a completely static image

Logically, the Samsung Galaxy S8 should not have problems with screen burn-in. The device is designed in such a way that it slowly but constantly changes the position of a static image on the display. For example, we are talking about a clock, it moves all the time. This information was recently confirmed on the Galaxy Club fan site. The manufacturer did this on purpose to avoid problems with burnout. The same update appeared on the Galaxy S7. Therefore, the current situation with display burnout on the Galaxy S8 is most likely simply a rare exception to the rule.

How Always On Display works

The Home button is part of the Always On Display, but it's not actually always on. It turns off when using certain apps and is even temporarily removed when you're on the unlock screen. Even if you use the display for a long time, for example, reading, the button will not work.

Based on this, the display will rarely create a static image over a long period of time. In fact, we still need to try to get the Galaxy S8 to do this.

Finally

No LED display is immune to burnout. Moreover, the company has taken special precautions for its new flagships. However, displays can still burn out. If you are worried about your S8/S8 Plus, simply turn off Always On Display or remove the navigation bar in settings to get rid of static images.