Tests of six video cards at first DX10 gaming benchmarks

Author: Dmitry Sofronov
Date: 26/06/2007

Half a year has passed since the time when Microsoft's new DirectX 10 API was released together with Windows Vista. Up to the recent time, we've had to be content with merely technology demos developed by video card manufacturers, but anyway the situation is gradually changing.

Today, we are presenting first tests in gaming benchmarks developed for DirectX 10. Quite recently, the 1.7.0.0 patch to Company of Heroes was released, which allows for running DX10 shaders on the existing engine. The second test is also based on the existing game Call of Juarez but is made as a standalone application and is not a patch enhancing the game's functionality.

As you may have guessed already, this is Radeon HD2900XT - the only available video card by AMD which offers support for DirectX 10. The specifications are well known to you – operating frequencies 740/1650 MHz, 512 MB memory, 512-bit memory bus width. For details of the architecture and special features of Radeon HD2900XT, read our review.

ASUS EN8800GTX AquaTank is also well known to you. The nominal frequencies of this video card are somehow higher than the those recommended for typical video cards of the 8800GTX class. To get an idea of performance of the "typical" 8800GTX in this test session, we had to push the frequencies of ASUS EN8800GTX AquaTank to the recommended values – 575/1800 MHz (engineers at ASUS, please forgive us).

The next in the "Table of ranks" of NVIDIA video cards goes GeForce 8800GTS with 640 MB video memory. A video card made by Chaintech Walton Corp. acts as a typical representative of the series. The specifications of the video card are standard – operating frequencies 500/1600 MHz, 640 MB memory, 320-bit memory bus width.

Since 8800GTS 640 MB is taking part in the tests, it would be a good idea to compare it against the version equipped with merely 320 MB of video memory. We have also got a video card like that made by ASUS. Apart from twice as less the video memory capacity, it makes no difference from its "elder sister". The frequencies are 500/1600 MHz, and the memory bus width is 320 bit.

But the "top-end" is not all. It is even more interesting to find out about the performance of mid-end video cards at DirectX 10 applications. That is why we included two GeForce 8600GTS video cards into the tests. What you can see above is made by Palit. The frequencies of the video card are standard for 8600GTS and amount to 675/2000 MHz, and 128-bit memory bus width. The key trait of this video card is 512 MB memory capacity, which is twice as much as it should be for a "typical" 8600GTS.

The most "typical" video card in today's tests there acts 8600GTS made by ECS. It offers frequencies 675/2000 MHz, the memory bus width 128 bit, and the video memory capacity 256 MB.

Tests

For the convenience of comparison of video cards, their results are marked with different colors - bright red stands for Radeon HD2900XT, of course. GeForce 8800GTX – marked in green. The results for 8800GTS 640/320 MB are highlighted in yellow/golden colors, respectively. The results for 8600GTS 512/256 MB – in cyan tints. The darker tint stands for 8600GTS with 256 MB memory capacity.

Tests in Company of Heroes ? DirectX 9.0c

Since most of the video cards in this test fall within the high-end class, we decided to set the graphics quality close to the maximum in the game settings. Below you can see a screenshot with the test parameters.

Of these settings, merely three options have undergone changes. The first option, Shader Quality, is in charge of the shader types used in the game. For tests in DirectX 9.0c, we set the option to "High". For tests in DirectX 10, we set the option to "Direct3D 10", respectively. The remaining two options which we modified are texture details (Low/High), and the FSAA mode (None/4AA). That is done to make it possible to estimate the effect of onboard video memory capacity installed on the video card and the extent of performance drop with the FSAA mode enabled. Let's see what we got in the end.

Amazingly, but at this relatively easy mode the Radeon HD2900XT not only manages to overtake its direct competitor 8800GTS but even 8800GTX. As regards the pairs 8800GTS 640/320 and 8600GTS 512/256, the performance differences in this mode are minimum. To all appearances, with low texture details there is no need for a large video memory capacity, and in the remaining aspects the specifications of video cards in these pairs are the same. Unfortunately, RivaTuner does not display the video memory usage under Windows Vista 64, so we have to assume only. Now let's set the texture details to "High".

In the absolute expression, the results have gone down a bit, which is natural. Radeon HD2900XT is still competing on par against 8800GTX. Increase in the texture details has not affected the relative performance for the pair 8800GTS 640/320, but the results for 8600GTS 512/256 have undergone a noticeable stratification. In fact, it was known earlier that 512 MB of video memory would not be redundant at all for "Company of Heroes". Now let's make the task more complicated through enabling FSAA 4x.

In this mode, Radeon HD2900XT is no longer capable of overtaking 8800GTX, nevertheless the advantage over 8800GTS is quite substantial. The pair 8800GTS 640/320 is still demonstrating an ideal synchronism of results. Stability is preserved also for the pair 8600GTS 512/256, but it is of some different kind. The version of the video card of greater memory capacity is almost 1.5 times faster than the "standard" 8600GTS. But that is enough only to provide comfortable FPS at resolutions no higher than 1024?768.

Tests at Company of Heroes in DirectX 10

Then, there follow the test results using DirectX 10 shaders. The remaining options left unchanged. As before, we start with the "easy" mode – low texture details with no FSAA.

Calm down and don't fall off your chair. Enabling DirectX 10 shaders has resulted in a more trustworthy and realistic picture but causes an almost twofold drop of results as compared to DirectX 9.0c in the absolute expression. As regards the relative performance, GeForce 8800GTX proves to be beyond competition at low resolutions. Radeon 2900XT demonstrates a performance comparable to 8800GTS, but with the rise of resolution its results do not drop so essentially, which allows it to be comparable versus 8800GTX at 1600?1200. In the pair of video cards 8600GTS 512/256, the results differ by approximately 2.5 times, and even at 1024?768 the comfortable FPS is already out of the question.

The picture that you have seen above does not change with the rise of texture details. GeForce 8800GTX remains an irrefutable leader at all the resolutions. Perhaps the installed onboard 512 MB of video memory is no longer enough for the Radeon. We'll find out if that is true once we get a chance to test a video card with 1 GB of onboard video memory. On the other hand, in this case there must be a difference in results for 8800GTS having different memory capacities, but for some reason we can't see that. Perhaps it is an abnormal result for this test. As regards the results for 8600GTS, then starting with 1280?1024 the video memory capacity stops being of importance. There is another argument in favor of hypothesis on increased requirements of the test to the video memory capacity (over 512 MB).

Enabling the FSAA results in further "lag" of results , but on the whole the picture does not change. Except the results for the pair 8800GTS 640/320 at 1600?1200 where the 640 MB version of video card demonstrates a 1.5 times superiority over the cut-down version. As regards the absolute results, even top-end video cards can't provide a comfortable FPS at resolutions higher than 1024?768. I wonder how many generations of DirectX 10 video cards should change before we can see FPS in this test comfortable for mid-end video cards at least at 1280?1024.

Perhaps the low results in this DirectX 10 test are cased by the test itself and not by the hardware? Let's see what the participants of our test will show at Call of Juarez DX 10.

Tests in Call of Juarez DX10 Demo

As we already stated, this benchmark is made as a standalone application, with a special launcher used to start the game and control the graphic settings. Below you can see a screenshot with pre-set "Low" quality settings.

Of the options available to modify, there is only the screen resolution and the details level, The "Advanced Settings" section displays detailed graphics quality settings which conform to the selected "Details" level. We run the first test at the minimum settings.

At these settings, Radeon HD2900XT takes a lead, with 8800GTX lagging behind just a little bit. The pair 8800GTS 640/320 is still demonstrating a stable parity of results. But for the pair 8600GTS 512/256 the results immediately separated. The 8600GTS 512 MB version looks more attractive, although it does not demonstrate a comfortable FPS.

The "Balanced" level of details differs from the minimum settings in that simple shadows of objects and FSAA 2? are enabled.

At that as well, Radeon HD2900XT performs not bad, but at 1600?1200 it anyway proves to be inferior to 8800GTX. The results for the pair 8800GTS 640/320 fall apart because the 640 MB version appears to be more attractive, which is natural. However sad it is, we have to admit that the performance of the regular 8600GTS drops well below the acceptable level. The 8600GTS 512 MB version demonstrates a threefold performance boost as compared to it, but that can be regarded as no more than an accelerated slide show.

The "High" graphics settings make use of even higher rate of FSAA and object shadows.

In the rivalry between Radeon HD2900XT and GeForce 8800GTX, the struggle is going on at a varying success. Substantial changes can be seen in the results for the pair 8800GTS 640/320. At 1600?1200, the 320 MB version sharply drops so that it even performs worse than the 8600GTS 512 MB version. The results for 8600GTS 256 MB don't need any comments - all is evident.

Let's try to find out what the cause of such sharp drops in results is – is it the shadow details level or the rate of FSAA? To this end, we'll do the following. We disable the FSAA in the details settings without changing any other options as it is shown on the screenshot.

Comparing this diagram with the previous one, we arrive at the conclusion it is anyway the FSAA which is in charge of the so sharp drop of results at the "High" details level. If we disable it, the drop of performance for 8800GTS 320 MB will vanish immediately. The results for Radeon HD2900XT will also go up substantially. Nevertheless, the performance of GeForce 8800GTX is affected by the presence/lack of FSAA in a much less extent. Quite possibly, enabling the FSAA results in increased consumption of video memory. But as we already stated above, it is still impossible to clarify this point at a sufficient certainty.

Final Words

Let's sum up the first totals. Comparison of video cards tested in DirectX 9 and DirectX 10 has shown that increase in realism results in a substantial drop of the FPS level. That's understandable - it is a tradeoff for the "nicety". The requirements to the video memory capacity are also high enough, even at relatively "easy" modes. Modern top-end video cards still don't demonstrate a smashing performance at DirectX 10, but they are able showing quite a comfortable FPS although not at the maximum settings. The performance of mid-end video cards still leaves much to be desired. All in all, it would not be wise to make far-reaching forecasts based on merely two benchmarks one of which is still a pure demo. On the other hand, games are not yet optimized for DirectX. Both game developers and video processor manufacturers still have a long way to go to make the new graphics an enjoyable experience. Let's wait and see.

Bus	PCI-Express
CPU	Intel Core2 Extreme Quad-Core QX6700 @ 2.93 GHz
MB	Foxconn N68S7AA-8EKRS2H
Memory	Corsair XMS Xpert DDRII-800 4x512 MB
OS	Windows Vista 64 Premium, DirectX 10
PSU	Hiper HPU-4M730 730 W