Conclusion on synthetic tests So debut of RightMark3D 2.0 for research on the site was held. Tests in a touch almost all aspects of innovation in Direct3D 10, they flexibly configured, allowing us to assess the relative performance of all the lines of Direct3D 10 chips from AMD and NVIDIA.
Both unified architecture of these companies have shown in our new Direct3D 10 tests is good, big failures were not found, except for a couple of cases with obvious errors in the drivers AMD.
Both family : R6xx G8x and high performance computing and texture, they are good at the complex components of all types. Taking the results in general, the decisions NVIDIA have certain advantages over the competition from AMD, for the moment, are ahead of their video, in most cases. But in some AMD chips tests showed better results, such as complex geometry tests and pixel shaders.
The advantage of AMD chips in such tests with increasing load even growing. So the outcome of the battle of DirectX 10 games has not been determined, it is rather difficult to say which of his rivals to win. It can only be assumed that there will be similar to our results in the majority of applications R6xx and G8x be close to each other would lead to some solutions on the basis of the company NVIDIA, AMD others. And this will depend largely on developers and their methods and algorithms.
Tests pixel shaders 4.0 showed that multiple texture with a relatively small sample loading ALU better job video NVIDIA. Solutions AMD, in turn, ahead of competitors in the computer tests pixel shaders. In one video card based on chip architecture R6xx showed very good results and ahead of competitors from NVIDIA operation and the situation in the second test is not yet clear because of errors in the drivers.
As we have noted, tests geometry and vertex shaders give different results, some leading solutions NVIDIA, AMD others. Because of the growing complexity of the work to go forward geometric Shader video AMD, it can be assumed that the applications with the extensive use of geometric shaders, if any in the near future, will lead the company's chips.
The last couple of tests RightMark3D 2.0-speed tests on the sample texture of vertex shaders. The figures shown in the results clearly indicate that the video card based on NVIDIA chips G8x perform our tests texture samples from vertex shaders faster than solutions based on the AMD architecture R6xx. This is different from the traditional balance between texture and computing chips from two competing companies. "whittle down" the number sheydernyh blocks and blocks TMU ROP quite hitting decisions middle and lower levels, greatly reducing their productivity.
Low card behind the masthead of times, the best of mid-end 2-3 times (on HD 2900 XT and GeForce 8800 GTS), and low-end further to 4-8 times. As evidenced by the results of game tests, has only Direct3D 9. Judging from the reactions we have results, the drivers for Vista AMD clearly worse refined than the drivers NVIDIA.
If the second company's products in our tests, no errors have been found, the decisions of AMD were two distinct issues : the rest of the line chips in the second "test" computer pixel shaders ( "Fire") and the head of the 2900 XT HD solutions in the most difficult test mode, the speed of the sample texture vertex shaders "Earth". It would like to believe that these shortcomings will be addressed in the next version of CATALYST drivers.