by Alexey Stepin , Anton Shilov, Ilya Gavrichenkov
02/16/2006 | 08:31 PM
Futuremark’s benchmarking suites of the 3DMark series are well known to any person who’s interested in consumer 3D graphics hardware. This suite is a powerful tool for getting detailed information about the performance of a modern graphics card and the developer is working hard to make it more objective and indicative of the real-life gaming experience. As a result, the 3DMark series has become an industry standard de facto. The different versions of the suite enjoy popularity among overclockers who record their achievements using 3DMark as well as among major graphics card and PC manufacturers who need advanced tests that make use of all the progressive technologies available at the current moment.
Graphics processors are evolving at a faster rate than any other PC component and benchmarking suites are becoming out-dated and inadequate just too rapidly. This is also the consequence of changes in the approach game developers take to creating their new products. As a simple example, pixel shaders are employed in games widely today and their complexity is higher than before to achieve a more realistic image, so older versions of 3DMark can’t correctly estimate the performance of a graphics card in such games. Furthermore, a benchmarking suite that aspires to be an industry standard must match not even today’s, but tomorrow’s games and serve to predict to some extent the performance of today’s graphics hardware in unreleased-yet next-generation applications.
As we noted in our 3DMark05 review, the 2003 version of the suite, 3DMark03, met the mentioned requirement by using stencil shadows and complex pixel shaders. It predicted quite well the performance of graphic cards in such popular games as Far Cry, Doom 3, Half-Life 2 and others. 3DMark05 came out next, stressing the use of a great number of pixel shaders in a single scene, i.e. exactly like in many today’s games like F.E.A.R., Battlefield 2, Splinter Cell: Chaos Theory, Call of Duty 2, Serious Sam 2 to mention but a few. This version of the benchmark can use Shader Model 2.0, 2.0a, 2.0b and 3.0 modes, but all its shaders, with rare exceptions, are limited to Model 2.0; the other modes can only affect speed rather than image quality. It means 3DMark05 is still a working tool when it comes to compare today’s graphics cards, but it does not suit for testing next-generation graphics hardware since it doesn’t have specific Shader Model 3.0 tests, doesn’t support HDR and is generally too simple to give a clear picture of how graphics cards will perform in games that will appear in the next 12-18 months. A new, tomorrow-oriented version of the benchmark was called for and it was developed. So, it is about 3DMark06 that we are going to talk today.
Before talking about the core concept of the new benchmarking suite, we’d like to give you a brief account of the previous versions of 3DMark and their ideology:
As opposed to the previous versions of the suite, 3DMark06 is not full of innovations. It can rather be considered as an update to 3DMark05, especially since three out of the four graphics tests of the new suite are nothing else but improved versions of 3DMark05’s gaming tests. So, the new version differs from the previous one quantitatively rather than qualitatively.
Really new in the new version of the suite are HDR support, Uniform Shadow Maps, multi-CPU configurations support, and the general orientation towards Shader Model 3.0, although two out of the four graphics tests are limited to Shader Model 2.0.
The rest of the changes are quantitative: the level of detail is once again increased in the test scenes, there are more light sources, the shaders are more complex, the textures have a higher resolution, etc. So, the main concept of 3DMark06 is to focus on SM3.0-compatible graphics processors.
As you probably know, a completely new graphics engine was developed for 3DMark05. It had nothing in common with the previously used MAX-FX engine and was more like an engine from a real-life game.
The 3DMark06 engine is its modification with an addition of such features as full support of Shader Model 3.0 and textures and blending in FP16 format. The latter means nothing else but High Dynamic Range – HDR. Futuremark predicts that HDR will be supported widely in games to come, although currently there are few such titles available. Like in 3DMark05, shaders that stand for a particular material are generated dynamically in HLSL format. Then they are compiled to match optimally the graphics processor of the system either automatically or according to a user-defined profile.
Textures and blending in FP16 format are only required by the SM3.0 graphical tests. These tests also use FP16 filtering, but if the GPU doesn’t support this feature, it is emulated with a special shader. This permits graphics cards with the Radeon X1000 architecture that can’t filter floating-point textures to pass the SM3.0/HDR tests. The SM3.0/HDR graphics tests also make use of post-processing to apply the Bloom effect, the “star” effect that emulates the 6-petal camera shutter, and the lens flare effect. And finally the image undergoes tone-mapping for correct color representation on traditional displays.
The developer says the new benchmark makes use of all the key features of SM3.0, except for the vFace register:
Dynamic shadows appeared in Futuremark’s benchmarks starting from 3DMark2001. They were created by means of projection shadow maps then. It was quite a simple method but with certain limitations like an object couldn’t cast its shadow on itself. Moreover, the shadow was projected on all the surfaces under the object, even on the floor of the room a few stories below. 3DMark03 introduced the so-called stencil shadows technique to create dynamic shadows. This method works differently: the boundaries of an object visible from the light source direction are selected as a polygon without lighting and everything within the volume of this polygon is shadowed. This method is free from the drawbacks of the previous one and allows an object to cast its shadow on itself, but it is not universal and suits only for certain types of scenes and low-polygonal objects.
However, the fact is the selection of the object boundary that becomes the volume shadow is a rather resource-consuming operation and polygons that comprise these shadows consume a large share of the total scene fill rate, even though they are invisible.
3DMark05 introduced a new method of generating dynamic shadows, using the so-called LiSPSMs (Light Space Perspective Shadow Maps). 3DMark06 improves this technique further by using another type of shadows maps, called Cascaded Shadow Maps or CSMs. Using CSMs allows drawing shadows for all the objects on the screen irrespective of their angle of inclination.
This method divides the view frustum in 5 sections along the Z axis. Each section is then shadowed using a standard uniform shadow map with 2048x2048 resolution. If the GPU supports depth textures, a depth map in D24X8 or DF24 format is used. Otherwise a component of an R32F texture (floating-point 32-bit representation) is used as a depth map. Hardware shadowing is on by default (except for D24X8 in the SM3.0/HDR tests), but the user can turn it off.
Each method has its downsides. Although the resolution of depth maps is high, it may sometimes prove to be not high enough and, like in 3DMark05, flickering may occur on the boundary of the shadow, the so-called projection aliasing. This thing may happen when the direction of the normal maps is strictly or near perpendicular to the direction of lighting. It is currently next to impossible to avoid this effect without serious performance loss.
A 16-sample array (4x4) is used in the SM3.0/HDR tests to smooth out the edges of shadows. This array is rotated randomly for each pixel of the shadow boundary. Having 16 reference points improves the antialiasing quality but requires additional computational resources. Per-pixel sampling is used for hardware shadow mapping as well as for shadow maps in R32F format. In the SM2.0 tests a smaller, 2x2-pixel array is used, but if the GPU can only sample from the depth buffer in formats D24X8, DF24 or Fetch4, one bilinear sampling is only made. The antialiasing quality differs at that somewhat. In case the user wants to compare the rendering performance of different architectures, hardware shadow mapping may be disabled and dynamic shadows are then created using R32F depth maps and are anti-aliased using 4 samples.
Generating dynamic shadows with depth maps is quite appropriate in 3DMark06. According to Futuremark, this method is already used by game developers and will be getting ever more popular in the future. As for texture compression, all color maps in 3DMark06 are compressed with the DXT1 algorithm, alpha-maps with the DXT3 algorithm, and normal maps with the DXT5 algorithm. The 3Dc method, specific for ATI Radeon X700 and higher cards, is not supported.
There are four graphics tests in the new 3DMark that fall in two groups, one of which is limited to SM2.0 and the other needs an SM3.0-compatible graphics card. We’ll describe the tests starting from the SM2.0 ones:
The next two tests use the SM3.0 profile exclusively and thus require a graphics card with Shader Model 3.0 support to run.
A special feature of 3DMark06 is the new ideology of overall score calculation. While the previous version of the benchmark used to produce its total score basing on the performance of the graphics subsystem alone, 3DMark06 calculates its scores using the results of the graphics as well as the CPU tests. In other words, the final score the benchmark produces depends on both the graphics subsystem and central processor performance.
This has been implemented because the developer wanted to make 3DMark06 a benchmark that would give a general estimate of the performance of a platform at large as concerns running contemporary 3D games as opposed to a benchmark that measures the performance of graphics subsystems relative to each other. This approach is well grounded because modern gaming applications call for a high-performing central processor and are expected to get more demanding in the future as developers pay more attention to better modeling of the physics and AI of in-game objects.
As a result, the CPU test is an integral and important part of 3DMark06 and it has been made closer to real life than before. The CPU test from 3DMark05 had nothing to do with games. The performance was measured using academic algorithms that were not used in real applications. For example, the CPU was executing vertex shaders, but is it an ordinary gaming task for it?
The problem with the CPU performance tests in the previous versions of 3DMark was that they didn’t use specialized algorithms like those employed in real-life games. The new 3DMark06 corrects this issue as the algorithms from its CPU tests are the same as the CPU performs in real 3D games.
The CPU performance is measured in 3DMark06 by modeling a real-life gaming situation the benchmark designers called Red Valley. The test shows a fortress squeezed in between two cliffs. The bottom of the cliffs is all in ravines speedsters are racing along. Each speedster is trying to avoid crushes and the fortress defense and to get inside the fortress. The defense is using some kind of flying tanks, slow but equipped with rockets. All in all, there are as many as 87 bots of these two types in the Red Valley scene.
It is the graphics subsystem only that is responsible for graphics output during the CPU test. To minimize the influence of the graphics subsystem on the CPU performance result, the test is performed in 640x480 resolution and dynamic shadows are disabled. The processor is doing its typical job. It processes the in-game logic, models the physics, and gives the bots their AI. The physics of the Red Valley scene is calculated using the AGEIA PhysX library, popular among game developers. The intelligence of the bots is the result of solving the path-finding problem.
With so many intelligent bots in the scene, the CPU test looks more like a real-time strategy, but you should be aware that 3DMark06 is not meant to resemble today’s games. It must predict tomorrow’s ones which, according to Futuremark, are going to have much more active and intelligent objects than today’s.
Being oriented at tomorrow’s games, 3DMark06 was optimized for today’s advanced dual-core processors. This test can also effectively load CPUs with more cores since the task of finding optimal paths for numerous objects can be easily paralleled. The calculations in the CPU test are divided into threads as follows: one thread calculates the game logic and controls the calculation process; the second thread is used to model the environment physics; the other threads are busy finding the optimal paths (there can be as many of them as there are execution cores in the system).
When testing processors in 3DMark06, the Red Valley scene is used two times with different algorithms settings. The first time more resources are allocated to model the artificial intelligence and the second time the environment physics is in the focus of the test.
3DMark06 incorporates all the theoretical tests from 3DMark05 as well as two new tests, Shader Particles Test (SM3.0) and Perlin Noise (SM3.0). As the names suggest these two tests both require Shader Model 3.0 support to run.
The rest of the tests, including the batch size tests, have been left the same. So, this is the end of the theoretical part of this review. We will now proceed to test contemporary graphics cards and CPUs in the new benchmark from Futuremark.
We investigated the performance of contemporary graphics cards in the new version of 3DMark on the following platform:
Ati, Nvidia and S3 drivers were set in the following way:
N vidia ForceWare:
For the overall performance results we ran 3DMark06 with the default settings:
In all other cases we used the corresponding FSAA and anisotropic filtering settings. Altogether we tested 20 graphics cards, which we split into three groups: high-performance, mainstream and value.
High-Performance Graphics Cards
Mainstream Graphics Cards:
Value Graphics Cards:
The CPU testing was conducted in platforms built with the following components:
We tested 19 CPUs from AMD and Intel. Here is the complete list of processors participating in our test session:
Now that we’ve learned how the CPU performance test works in 3DMark06, it’s time to do some practical testing. First, here are the overall CPU performance scores that 3DMark06 produces:
The first thing we should mention is that the CPU test from 3DMark06 is really well optimized for dual-core processors. The difference between the single- and dual-core CPUs strikes your eyes immediately.
Among dual-core processors, the models from Intel’s Pentium D and AMD’s Athlon 64 X2 series keep on roughly the same level. Although AMD’s processors are a little ahead of Intel’s processors of the same price, the gap is a negligible 5%.
Intel’s Pentium 4 is superior among single-core processors, most probably due to Hyper-Threading technology.
If you are interested in details, take a look at the results of both the subtests the overall CPU rating is calculated in 3DMark06.
Although these tests vary in their use of physical modeling and AI algorithms, the relative results are almost the same.
And lastly let’s see how strongly the CPU performance affects the overall score in 3DMark06.
So, the CPU does affect the overall score, but not too much. The weight coefficients in the overall score calculation formula are selected such that it is the graphics subsystem performance that becomes the basis for a 3DMark06 score.
3DMark06 demands fast execution of an abundance of complex shaders. This is why the Radeon X1800 XT can’t beat the GeForce 7800 GTX 512 and even loses to the GeForce 7800 GTX which compensates its rather low GPU and memory frequencies with its 24 pixel processors. The inability of the Radeon X1000 architecture to filter FP16 textures probably affects the results, too. Using an additional shader for “software” filtering of such textures may have created an extra load on the GPU and reduced the performance of the card.
The new ATI Radeon X1900 family, on the other hand, has a higher computational capacity than the Radeon X1800 series and proves it in this test. For example, the senior Radeon X1900 XTX model is quite confidently ahead of the previous leader. The less advanced model has a smaller advantage, yet it is a little better than the GeForce 7800 GTX 512 thanks to its 48 pixel processors as well as to the Fetch4 feature that accelerates processing of dynamic shadows created by means of shadow mapping. We should expect a more impressive victory from the Radeon X1900 series in the SM3.0 graphics tests.
In the lower category, the Radeon X1800 XL also loses to the GeForce 7800 GT while the last place goes to the out-dated and obsolete Radeon X850 XT Platinum Edition that doesn’t support Shader Model 3.0.
The overall result of the two graphical tests isn’t good for the Radeon X1800 family even though the tests don’t use HDR. Why? These graphics cards have only 16 pixel processors and the Radeon X1000 architecture is generally less efficient when there are numerous simple shaders to be processed. Unlike them the Radeon X1900 cards feel much more confident in the SM2.0 graphical test. Even if they can’t work as efficiently with version 2.0 pixel shaders as Nvidia’s GeForce 7800 GTX 512, they make up for it with their numerous pixel processors, high core frequencies, Fetch4 support and the larger hierarchal Z-buffer.
Not so long ago the second round would have been Nvidia’s, but the recently announced Radeon X1900 XTX snatches the victory away of the GeForce 7800 GTX 512 while the cheaper Radeon X1900 XT has the same result as the Nvidia card’s up to 1 point. Moreover, the newcomer from Ati is much more affordable: you can already buy it today, while GeForce 7800 GTX 512 is still pretty hard to find and costs the whole lot of money.
The superiority of the Radeon X1900 family in the SM3.0/HDR tests appeared not so overwhelming as we have anticipated. Nevertheless, even the Radeon X1900 XT currently selling for $549 is somewhat faster than the GeForce 7800 GTX 512, which is most likely to cost you well over $700 if you are lucky to find it at all.
Despite its architectural innovations, the Radeon X1800 XT is in fact out of competition although it should have had an advantage in the third and fourth tests. Yes, the GeForce 7800 GTX 512 with its 24 pixel processors clocked at 550MHz and 1.7GHz graphics memory just has more raw power, but how do you explain the Radeon X1800 XT’s being as slow as the ordinary GeForce 7800 GTX which has much more modest technical characteristics?
So, the ATI Radeon X1900 XTX now enjoys leadership in the high-performance graphics card sector, which belonged to the Nvidia GeForce 7800 GTX 512 just yesterday, but the Radeon X1800 performs rather poorly. It’s unlikely that we will find any changes in the rankings after analyzing the results of the graphical tests one by one, but let’s still try to do so.
The GeForce 7800 GTX is always ahead of the Radeon X1800 XT, but by no more than 1fps or less than 10%. The Radeon X1800 XL does worse. It is about 15% slower than the GeForce 7800 GT.
The Radeon X1800 family is no rival to the GeForce 7800 GTX 512, but the new Radeon X1900 allow the senior graphics card from Nvidia to win the lowest resolution only, 1024x768. In higher resolutions the Radeon X1900 XTX is unrivalled and the Radeon X1900 XT keeps on the same level with Nvidia’s fastest product.
In 1280x1024 with enabled full-screen antialiasing and anisotropic filtering, the Radeon X1800 XT at last manages to outpace the GeForce 7800 GTX a little, and the Radeon X1800 XL to get even with the GeForce 7800 GT. The GeForce 7800 GTX 512 is only challenged by the Radeon X1900 XTX and, in high resolutions, by the Radeon X1900 XT, too.
Nvidia’s graphics cards, except for the senior model, suffer from graphics memory misbalance problems in 1600x1200 and can’t pass the test whereas the Radeon X1800 XL does, although is equipped with exactly the same amount of graphics memory, 256MB.
The gap between the Radeon X1800 XT and the GeForce 7800 GTX diminishes almost to zero in the second test. In 1600x1200 resolution it is no bigger than 0.4fps, for example. This scene requires high math1ematical performance from both pixel and vertex processors while the fill rate is less important due to the small size of the scene.
The GeForce 7800 GTX 512 is still ahead of the Radeon X1800 XT, but not quite confidently. The Radeon X1900 XTX pushes it down from the top position altogether. This test isn’t ideal for the new series of graphics cards from ATI, but the complex lighting and shadowing of the scene is where the Fetch4 feature supported by the Radeon X1900 cards can prove its efficiency.
The Radeon X1800 and X1900 series cards profit by the ring-bus memory controller when full-screen antialiasing is turned on. Here, the Radeon X1800 XT can’t beat the GeForce 7800 GTX 512, but is a mere 10% behind it in 1600x1200 resolution. The Radeon X1800 XL overtakes the GeForce 7800 GTX, leaving the GeForce 7800 GT behind in 1280x1024 and higher resolutions.
As for the Radeon X1900 XTX, it enjoys an even heftier advantage than in the pure speed mode. Despite the small difference in the clock rates, the Radeon X1900 XT is noticeably slower than the senior model exactly when FSAA and anisotropic filtering are on, although it delivers enough performance to catch up with the GeForce 7800 GTX 512.
Once a leader of the race, the Radeon X850 XT Platinum Edition now has to default as it does not support Shader Model 3.0. The standings are quite different from what we’ve seen in the SM2.0 tests: the Radeon X1800 XT isn’t quite unrivalled, yet is ahead of the GeForce 7800 GTX nonetheless. Its results can be compared to those of the GeForce 7800 GTX 512 although the latter is a little faster due to its impressive technical characteristics.
Meanwhile, it is the first SM3.0 graphics test with a lot of long and complex shaders that puts the advantages of ATI’s new concept – the number of pixel processors is three times the number of TMUs – in the spotlight. The Radeon X1900 XT is 20% faster than the GeForce 7800 GTX 512 across all the resolutions! As for the Radeon X1900 XTX, this flagship product from ATI Technologies just confidently holds its first place since the very beginning.
We didn’t test the “eye candy” mode for two reasons. First, it’s only the ATI Radeon X1000 family that allows using FSAA and HDR simultaneously, so there would be no comparisons to make. And second, in tomorrow’s games the Radeon X1800 XT and even the Radeon X1900 XTX are going to be too low-performing to deliver a playable frame rate in high enough resolutions with enabled HDR and full-screen antialiasing.
The second SM3.0 graphical test makes use of a number of complex shaders, thus singling out GPUs that have many pixel processors efficient at executing long shaders. The Radeon X1800 XT is again a little behind the GeForce 7800 GTX and the Radeon X1800 series can’t match the performance of the GeForce 7800 GTX 512. The newer Radeon X1900 XT can do that, however, as it makes good use of its tripled number of pixel processors and Fetch4 support. Of course, the senior Radeon X1900 XT model is steadfastly on the top.
So this test session shows that the Radeon X1900 XTX is the absolute leader in 3DMark06 among single graphics cards. The second place is shared by the GeForce 7800 GTX 512 and Radeon X1900 XT which match each other’s performance with a minor advantage on either side in particular tests. The exception is the second SM3.0/HDR test that evidently prefers graphics cards with high math1ematical performance.
Having a progressive architecture, the Radeon X1800 XT still has to content itself with the fourth place, yielding the third to the GeForce 7800 GTX. The gap is small, however, so these two cards can be viewed as similar. The Radeon X1800 XL on its part couldn’t match the GeForce 7800 GT due to the low GPU clock rate. 500MHz proved to be too low to make up for the opponent’s having more pixel processors which are also more efficient when there are many texture fetches to be made.
ATI’s mainstream graphics cards of the Radeon X800 generation fail in 3DMark06, but the new Radeon X1600 XT with its TMU-to-pixel processor ratio of 1/3 looks quite competitive against the GeForce 6800 GS. The idea of ATI Technologies to increase the number of pixel processors and improve their performance finds approval in the new benchmark from Futuremark. Considering that 3DMark is always oriented at next-generation games, and we already can see the tendency to use more arithmetically complex pixel shaders in games, increasing the number of pixel processors seems to be the optimal strategy to build new GPUs today. Back to the results of the tests, the Radeon X800 XL managed to outperform the GeForce 6800 which was hamstringed by its low GPU and memory frequencies.
The GeForce 6800 looks poor in the SM2.0 graphical tests, but these tests become the winning ground for the GeForce 6800 GS that leaves the 16-pipelined Radeon X800 XL behind. The Radeon X1600 XT is nearly as fast as the Radeon X800 XL. It doesn’t have a very high fill rate and it doesn’t work fast in the framework of Shader Model 2.0, while complex version 3.0 shaders are missing in the first two graphics tests of 3DMark06.
The Radeon X800 cards have to leave the race as they don’t support SM3.0. The GeForce 6800 definitely plays in a lower league than the three remaining products (GeForce 6800 GS, GeForce 6800 GT and Radeon X1600) which struggle desperately to win this test. The Radeon X1600 XT comes out on top as it was specifically designed for pixel shader-heavy applications. It is followed quite closely by the GeForce 6800 GS, which scores a mere 38 points less. The GeForce 6800 GT is slower due to the low frequency of the GPU. After all, the NV40 chip will be 2 years old in the next April and the NV45 differs from it only in having a HSI bridge integrated into the die packaging.
Taken independently, the results of the first SM2.0 test look quite right: the GeForce 6800 GS and 6800 GT are in the lead, but the Radeon X800 XL closes the gap in higher resolutions due to the more efficient memory controller. The Radeon X1600 XT isn’t any slower than the Radeon X800 XL: its architecture may not be the most suitable for this particular test scene, but it does have high GPU and memory frequencies to make up for that. Lower in the standings table is the Radeon X800 GTO. The GeForce 6800 can’t pass the test in 1600x1200 resolution even with disabled FSAA – 128 megabytes of memory is insufficient for that.
Like their high-end mates, Nvidia’s mainstream graphics cards can’t pass the test in 1600x1200 resolution at the “eye candy” settings, and the GeForce 6800 even refuses to do so in 1024x768. Anyway, the GeForce 6800 GS is still in the lead in lower resolutions.
The second SM2.0 test tells us the same things as the first one, except that the speeds of the cards are higher here because of the smaller size of the scene. Like in the previous case, the GeForce 6800 GS wins.
Here’s a surprise as we switch into 4x FSAA + 16x AF mode: the GeForce 6800 GS was the leader at these settings in the first test, but now it loses its top place to ATI’s Radeon X1600 XT and Radeon X800 XL which deliver near identical frame rates, up to 0.2fps precision. This is all rather strange as the test doesn’t require high graphics memory performance, but rather fast and efficient processing of vertex shaders and dynamic shadows. This explains the failure of the GeForce 6800 GT, however, that clocks its GPU at 350MHz only.
The first SM3.0 graphical test is the most difficult of all. Besides version 3.0 shaders, it uses HDR, so the Radeon X1600 XT gets a chance to show its talent and it does so in all the resolutions. However, in 1600x1200 resolution the GeForce 6800 GS overtakes the Radeon X1600 XT – the difference of 0.1fps is within the measurement accuracy range. This is curious, but bears no harm to the reputation of the Radeon X1600 XT in our eyes because this graphics card is not meant for high resolutions and the GeForce 6800 GS too, for that matter. Don’t forget we’re talking about tomorrow’s games where both these cards will most likely deliver a more or less acceptable performance in low resolutions only.
In the last graphical test of 3DMark06 the main opponents in the mainstream sector perform in a similar way in low resolutions, but the GeForce 6800 GS breaks away from the Radeon X1600 XT starting from 1280x1024. And we have to repeat it once again: high resolutions will be unavailable in tomorrow’s games for owners of today’s mainstream graphics cards.
So, the mainstream sector doesn’t have an all-around leader as the top-end sector has. The GeForce 7800 GTX 512 is obviously the best among highest-performing solutions, but with mainstream products it all depends on the type and complexity of shaders employed. The SM2.0 tests prefer the GeForce 6800 GS almost always, even when FSAA is in use, while the more difficult SM3.0/HDR tests run generally faster on the Radeon X1600 XT, the embodiment of the concept “max performance at executing math1ematics-heavy pixel shaders”. In other words, the GeForce 6800 GS is the best choice for today’s games in which Shader Model 3.0, not to mention HDR, are but seldom employed, while the Radeon X1600 XT is somewhat more future-proof.
The Value graphics cards seem to be the most numerous group in our today’s test session: there are 8 participants here, while the previous two groups had only 6 each. Here you can see such time-tested veterans as Radeon X700 PRO and GeForce 6600 GT alongside with the relatively new Radeon X1300 PRO and GeForce 6600 GDDR2 and the brand new S3 Chrome S27 256MB. The latter demonstrates surprisingly good result running neck and neck with Radeon X1300 PRO, even though it doesn’t support SM3.0. Only GeForce 6600 GT shows higher results than the others. Although it only uses 128bit memory bus, due to SM3.0/HDR support it manages to outperform Radeon X800 GT in the overall score chart, as the latter can only pass two graphics tests out of four. As in the previous cases, we cannot draw any conclusions basing on the total scores only, so let’s take a closer look at each given test individually.
As we have expected, S3 Chrome S27 256MB wins in SM2.0 graphics tests, because at this time it is the world’s only budget GPU working at 700MHz frequency. According to our results, it boasts pretty high fillrate and copes quite well even with complex pixel shaders 2.0 (for details see our S3 Chrome S27 Graphics Processor Review: Worthy Performance for Its Class? ). The next fastest in this bunch is Radeon X800 GT, which also feels quite at home in those benchmarks that do not use SM3.0, but GeForce 6600 GT is not that much slower either. Radeon X700 PRO/X700 have some caching issues and besides, they work at relatively low frequencies, so their results are pretty modest, I should say. However, the same is true for GeForce 6600, which suffers from relatively slow memory working at only 250 (500) MHz.
Ati’s orientation at the maximum performance with SM3.0 reveals itself positively in the value graphics cards segment. Although the technical specifications of the Radeon X1300 PRO are considerably more modest than those of the GeForce 6600 GT, the results of these two cards are almost identical: the difference makes only 14 points. It is an excellent result for a graphics card with only 4 pixel and 2 vertex processors. Now let’s take a look at the results of each graphics test separately.
S3 Chrome S27 256MB carries 4 vertex processors onboard that work at 700MHz. As a result it can compete on equal terms with such powerful rivals as Radeon X800 GT and GeForce 6600 GT. This is very impressive especially keeping in mind that all previous attempts of S3 Graphics to design a competitive solution ended in vain. Of course, they had to pay for their success: the new S3 solution features expensive GDDR3 memory with 1.4ns access time. Among other testing participants we can single out GeForce 6600 GDDR2 – a successful modification of GeForce 6600 that acquired fast memory and higher GPU frequency. Since the value graphics cards performance in 3DMark06 lies between 2 and 8fps even in pure mode, we decided it wouldn’t make any sense to run benchmarks in eye candy test mode at all.
In the second SM2.0 test the GPU frequency plays even bigger role than in the first test: the math1ematical performance of the pixel and vertex processors depends directly on the GPU frequency. The leadership here again belongs to S3 Chrome S27, which is closely followed by Radeon X800 GT due to more vertex processors it has. Although Radeon X1300 PRO cannot really find a task up to its abilities, it is nearly as fast as GeForce 6600 GDDR2 and Radeon X700 PRO.
The situation changes dramatically, when it comes to SM3.0/HDR graphics tests. Here Radeon X1300 PRO can only compete with GeForce 6600 GT, but just take a look at the price difference and the technical specifications of these two cards! Ati Radeon X1000 architecture once again proves that it is more cut for the future.
The second SM3.0 graphics test is simpler than the first one and GeForce 6600 GT regains its status quo. However, the second place is stably taken by the shy Radeon X1300 PRO, and the gap between the leader and the second best is really not that big: 10%-12% in 1280x1024.
Unlike the mainstream graphics cards, the winner in the value segment is indisputable: it is GeForce 6600 GT. It is at least as fast as Radeon X1300 PRO in SM3.0/HDR tests, and in SM2.0 it loses only to Radeon X800 GT and S3 Chrome S27 256MB. The standard modification of this graphics card is equipped with 128MB of graphics memory, but it is very unlikely to affect the performance in real games: you have to play in high resolutions to real feel short on video memory, and GeForce 6600 GT doesn’t have enough performance to provide comfortable gaming at high resolutions. This graphics card is a sort of a universal solution and suits very well for games using only SM2.0 as well as for games using SM3.0. On the other hand, Radeon X1300 PRO costs less and runs almost as fast in SM3.0 tests, therefore, it might be a better choice for those users who aim at the newest gaming titles but have limited budget for a graphics card upgrade. The S3 Chrome S27 256MB graphics card demonstrated truly phenomenal results here, however, it doesn’t have any future: it has no Shader Model 3.0 support, although it does really well in games using SM2.0.
Futuremark 3DMark testing suites have always looked into the future being at the progress forefront. Contemporary graphics cards have never shown any superior results in these tests, but it has always been the distinguishing feature of this test software. It was an excellent tool for predicting the future and revealing the major trends in 3D graphics world. 3DMark05 once predicted the wide-spreading of the Shader Model 2.0, and now it has already become a common thing for us: any contemporary game uses a lot of shaders 2.0. So, what does 3DMark06 tell us these days?
And in the meanwhile 3DMark06 can serve as a pretty good tool for testing graphics cards of different categories:
Radeon X1900 XTX: this graphics card shows the today’s best gaming performance and has quite a bit of potential for the future. The absence of FP16 textures filtering is no big drawback, as it can be easily simulated with the pixel shaders. With 48 pixel processors onboard, this emulation will not result into any noticeable performance drop.
GeForce 7800 GTX 512: good performance and good potential for the future. No simultaneous support for FSAA and HDR is no fatal thing: the performance would be too low in this mode anyway. It competes quite successfully with Radeon X1900 XT in terms of gaming performance, but cannot boast the same attractive price and wide availability.
GeForce 7800 GTX: this guy performs as fast as Radeon X1800 XT due to more pixel processors onboard. If the games use a lot of SM3.0/HDR it may yield to the latter that is why it is not as attractive on the long-term basis as Radeon X1800 XT.
GeForce 7800 GT: this solution runs as fast as Radeon X1800 XL, and represents the best choice among the inexpensive performance solutions. It can theoretically yield to the latter one in games using a lot of SM3.0 shaders, however, the SM3.0/HDR graphics tests do not prove this.
Radeon X1600 XT: it looks less attractive than GeForce 6800 GS in contemporary games, but seems to be more promising thanks to a more progressive architecture.
GeForce 6800: isn’t very prospective because of relatively low performance and small amount of video memory.
Radeon X850/X800 family has no future, because it doesn’t support SM3.0/HDR. However, if you are mostly playing gaming without these features, they might be a good choice.
Radeon X1300 PRO is the most promising graphics accelerator in the budget segment, just as GeForce 6600 GT. The latter however is more expensive. It will offer acceptable performance in the future games only in low resolutions, such as 800x600 or less. The same is true for GeForce 6600 GDDR2 and GeForce 6600.
Radeon X700 family has no future because of low gaming performance and no Shader Model 3.0 support.
S3 Chrome S27 offers pretty high performance for a value graphics solution. However, it has absolutely no future ahead, because it doesn’t support SM3.0.
As for the 3DMark06 itself, Futuremark keeps up the good tradition of offering quality testing tools. The benchmarking suite is very convenient to work with and supports all the latest technologies. Its set of tests is outstanding from the technical as well as aesthetical points of view. 3DMark06 is destined to follow into the footsteps of its predecessors: it will undoubtedly become a new standard for 3D industry.