Direct3D 10: vertex texture fetch rate

Vertex Texture Fetch tests measure the speed of many vertex texture fetches. These tests are essentially similar, and the correlation of their results in Earth and Waves tests must also be similar. Both tests use displacement mapping based on texture fetches. The only major difference is that the Waves test uses conditional branches, while the Earth test does not.

Let's analyze the first test (Earth). As we failed to enable the SLI mode in our review, we'll analyze two diagrams for both modes: "Effect detail Low" and "High" with more texture lookups:

Judging by our previous reviews, results in this test are affected by texturing speed, ROP speed, and memory bandwidth. Both diagrams show similar relative positions of the cards, GeForce GTX 295 is closer to GTX 280 SLI than to GTX 260 SLI. But the dual-GPU card from the RADEON HD 4800 series leads in all modes -- unlike NVIDIA cards, it works in the nominal mode. Judging by previous tests, multi-GPU rendering really proves its worth. It's a pity NVIDIA cards did not use it in our tests. So our comparison is not complete.

Let's take a look at results of the second vertex texture fetch test. The Waves test executes fewer texture lookups, but it uses conditional branches. The number of bilinear texture lookups in this case reaches 14 (Effect detail Low) or 24 (Effect detail High) per each vertex. Geometry complexity changes just like in the previous test.

Performance gap between HD 4870 X2 and NVIDIA cards grows even wider in the second test, because the Waves test favors AMD products, and new RADEON HD 4800 models look very good here. Even virtually doubled performance (SLI) won't help GeForce GTX 295 catch up with its competitor in heavy modes. What concerns a simple mode, affected by memory bandwidth, even single-GPU cards perform almost on a par with dual-GPU solutions. Performance differences between these two versions of the test are not big, only results of AMD cards improve versus NVIDIA cards, as the load grows.

Conclusions on the synthetic tests

Results of our synthetic tests confirm conclusions made in all previous articles devoted to multi-GPU rendering. As GT200b does not differ from GT200, and the dual-GPU GeForce GTX 295 card combines two such chips in SLI, we come up with a natural conclusion: this graphics card outperforms single-GPU cards from NVIDIA in all synthetic tests where AFR works. It ranks in between SLI systems based on GTX 260 and GTX 280, closer to the latter.

The most unexpected conclusion in this part is that the new GeForce GTX 295 has failed to outperform RADEON HD 4870 X2 in any of our synthetic tests. Some of the blame certainly goes to the single-GPU rendering mode, set by the NVIDIA drivers in our RightMark 2.0 for all SLI systems. If it hadn't been for this annoying defect, GTX 295 would have enjoyed victories in Direct3D 10 tests. Nevertheless, the old group of tests (Direct3D 9) revealed the weakness of all solutions based on GT200 -- their operating frequencies are too low to catch up and outperform their rivals in all tests.

But despite being defeated by the AMD card in all synthetic tests (for objective reasons), this dual-GT200b card is a robust solution with a lot of execution units and excellent theoretical parameters. The situation may change dramatically in games, and GeForce GTX 295 will defeat HD 4870 X2 in most cases. Results in synthetic D3D10 tests also hint at this assumption, where all SLI systems with NVIDIA chips used only one GPU, but they still performed well in all tasks, except for purely arithmetic tests.

So let's proceed to the next part of our review with the most interesting tests of the new dual-GPU solution from NVIDIA as well as other graphics cards in several modern games. We can assume that GeForce GTX 295 will outperform RADEON HD 4870 X2 in most games, provided SLI and AFR work fine there, except for rare occasions of insufficient local video memory and heavy modes like MSAA 8x, where NVIDIA cards do not perform very well.

Write a comment below. No registration needed!