3DMark Vantage Feature Tests

This benchmarking suite isn't new, but its Feature tests support D3D10. Those are generally interesting, because they differ from ours. Perhaps, we'll be able to draw new helpful conclusions not indicated by RightMark.

Feature Test 2: Color Fill

This one uses a simple pixel shader, which does not limit performance. An interpolated color value is written into the off-screen buffer (render target) with alpha blending. It uses a 16-bit off-screen buffer of the FP16 format, which is often used in games employing HDR, so this is a timely test.

The performance results demonstrated by this test do not correspond to what we've seen in similar benchmarks, even considering the difference in formats (we used an integer buffer, 8-bit per component, while 3DMark Vantage uses 16-bit floating-point). It seems 3DMark Vantage demonstrates the approximate memory bandwidth instead of ROP performance. As for dual-GPU graphics cards, GeForce GTX 295 shows worse results than expected.

These test results more or less correspond to the theory. They depend on memory bus, type and clock rate. GeForce GTX 285 shows a decent result thanks to its 512-bit memory, while GeForce GTX 480 doesn't outperform it much due to the moderate GDDR5 clock rate and the 384-bit bus. Radeon HD 5870 is also nearby, with its 256-bit bus and fast GDDR5 memory.

Despite the higher-bandwidth GDDR5 memory, GeForce GTX 480, as well as Radeon HD 5870, is only a bit faster than GeForce GTX 285 that has a 512-bit bus and GDDR3 memory. This may indicate a potential bottleneck in case FP16 rendering buffers are used -- something that is very widespread in today's games.

Feature Test 3: Parallax Occlusion Mapping

It's one of the most interesting feature tests, as this technique is already used in games. It draws one quadrangle (to be more exact, two triangles) using Parallax Occlusion Mapping that imitates complex geometry. The test uses resource-intensive operations to trace rays and high-res Z maps. This surface is also shaded using the heavy Strauss algorithm. This test uses a very complex and heavy-load pixel shader with multiple texture lookups during ray tracing, dynamic branches, and complex Strauss lighting algorithms.

This test is different, because it depends on shader power, branching efficiency, and texture fetch rate combined. That is, it takes a balanced GPU and card to reach high speed. The efficiency of executing branches in shaders has a strong effect on results as well.

Unfortunately, GeForce GTX 480 performs moderately in this test, just 23% faster than GeForce GTX 285. It also lags behind GeForce GTX 295, as well as Radeon HD 5870. Obviously, Radeon HD 5970 is still out of reach.

The reasons for such low results are not clear. Perhaps, we should blame the low speed of texture fetches that are actively used in this test. Because GF100 is quite efficient in terms of branching, as confirmed by PS 3.0 tests. NVIDIA solutions have always been efficient in this test, but now Radeon HD 5870 outperforms even the new GeForce GTX 480. Perhaps, GF100 will do better in physical simulation tests?

Feature Test 4: GPU Cloth

This test computes physical interactions (cloth simulation) using a GPU. It uses vertex simulation with the help of vertex and geometry shaders, with several passes. Stream out is used to move vertices from one pass to the other. Thus, this feature test benchmarks execution of vertex and geometry shaders, and stream out speed.

Disregard the results of dual-GPU cards, those are clearly similar to the results of the single-GPU solutions (each GPU in Radeon HD 5970 and GeForce GTX 295 works at a lower clock rate than those in Radeon HD 5870 and GeForce GTX 285). In this test rendering speed depends on geometry processing and geometry shaders performance. Even GeForce GTX 285 shows good results, just slightly yielding to Radeon HD 5870. In turn, GeForce GTX 480 shows its power again.

In this test GF100 is almost 2 times faster than the predecessor. This corresponds to its doubled shader power. The advantage over Radeon HD 5870 is as impressive. In other words, the novelty is obviously a leader in geometry shaders (and geometry processing in general) -- all according to the theoretical data.

Feature Test 5: GPU Particles

Physics simulation test based on particle systems computed with the help of GPUs. The test also uses vertex simulation, where each vertex is a single particle. Stream out is used for the same purpose as in the previous test. The test computes hundreds of thousands of particles, all of them being animated separately, and their collisions with a bump map. Similar to one of our tests in RightMark3D 2.0, particles are drawn with a geometry shader, which creates four vertices from each point and forms a particle from them. However, the heaviest load falls on shader units (vertex computing), stream out is used as well.

The result is even better. GF100 just grinds its rivals to dust in synthetic cloth and particle imitations of the Vantage suite that use geometry shaders. This time it outperforms the predecessor by almost 3 times, Radeon HD 5870 -- by about 2 times.

The dual-GPU cards show results similar to those of the previous test. Both cards obviously don't have AFR working -- the current frame data is used in the next frame, so the card can't start processing the latter before it has finished rendering the former. This is the drawback of dual-GPU solutions, they just can't work efficienly when next frames use data from previous frames.

Feature Test 6: Perlin Noise

The last feature test is arithmetically intensive for a GPU. It calculates several octaves of Perlin noise in a pixel shader. Each color channel uses its own noise function for higher GPU loads. Perlin noise is a standard algorithm, which is often used for procedural texturing, it's a mathematically complex procedure.

This math feature test demonstrates pure GPU performance in peak-load tasks. Its results correlate with theory and partially correspond to what we've seen in our own RightMark 2.0 math tests. However, this one indicates a bigger difference between the tested graphics cards.

So, GeForce GTX 480 finally outperforms GeForce GTX 285 by exactly 2 times. But it still lags behind Radeon HD 5870 too much, by about 1.7 times. Not to mention Radeon HD 5970.

In other words, AMD products beat NVIDIA's hollow in this test. Though GF100 has caught up a bit nevertheless. Just to remind you, this test is rather straightforward, called to demonstrate performance close to the theoretical peak. While more complex tasks, e.g. physical computing, produce different results, simple math is still done faster by AMD graphics cards.

Write a comment below. No registration needed!