Direct3D 10: Vertex Texture Fetch

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) in Effect detail Low mode:

Judging by our previous reviews, this test is equally affected by texturing speed and by memory bandwidth. And our today's results confirm it.

There is a very interesting difference between AMD cards. HD 5870 is the leader, but it's only a little faster than the HD 4870, which can be explained with its insignificantly increased memory bandwidth (the same GDDR5 memory and the 256-bit bus, only the frequency is higher.) Our today's cards under review failed this test because of their low memory bandwidth, they are up to 1.5 times as slow as their competitors from NVIDIA here. Let's take a look at results of this test with more texture lookups:

Relative positions of our contenders haven't changed much, all the cards demonstrate slightly worse results. The HD 5870 is ahead of the HD 4870 again, but not that much. They are followed by both GeForces. And the new HD 5700 solutions bring up the rear. Memory bandwidth is very important, but mostly in the easy mode. The heavier the mode, the better relative results are demonstrated by the new Juniper cards.

Let's have 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.

Waves test results resemble what we saw in the previous vertex texture fetch test. But now NVIDIA products have almost no advantage over RADEON HD 5700, it remains only in the easy and medium modes for the HD 5770 and GTX 260. The HD 5870 is far ahead, and the HD 4870 is a little outperformed.

The new RADEON HD 5700 models suffer from insufficient memory bandwidth in this synthetic test. But as we have already mentioned, it's enough to compete with the GeForce. Let's analyze the second modification of the test:

There are almost no changes, everything is the same. GeForce GTX 260 is again slightly ahead of RADEON HD 5770, and the HD 5750 performs on a par with the GTS 250. This time RADEON HD 5870 is even more than twice as fast as the HD 5770, which can be explained with minor architectural differences (cache size, their throughput, etc.) However, it looks like our VTF tests in RightMark 2.0 cannot show adequate differences between such solutions anymore.

Conclusions on the synthetic tests

Synthetic tests of the new RADEON HD 5770 and HD 5750 cards, based on RV840 Juniper, as well as other graphics cards from both chipmakers show us that the new solutions from AMD perform on a par with the previous top solutions, like RADEON HD 4870. That's a success. We can assume that their results in games will also be high, even in competition with NVIDIA.

The new GPU belongs to the RV8xx architecture that got support for DirectX 11 API. It's little different from the RV870 Cypress from the architectural point of view (except for the lack of double precision, which is not required in gaming graphics.) Its apparent advantage provided by the new 40nm process technology is a lot of execution units as well as a relatively high clock rate, accompanied by small dimensions and low manufacturing costs. So we can expect excellent results in games and a good price/performance ratio.

As we noticed in several synthetic tests, in some cases performance of Juniper-based solutions may be limited by memory bandwidth. This chip has a 128-bit memory bus, which is twice as narrow as the one in the RV770 and RV870. Fast GDDR5 memory helps raise memory bandwidth, but it may be insufficient in those cases, when render speed is limited by the effective fill rate. The new solutions may be sometimes even outperformed by the HD 4870 and HD 4890 in real games.

And as for the competition with NVIDIA, RADEON HD 5770 and HD 5750 demonstrate excellent results in most synthetic tests. They significantly outperform GeForces in all arithmetic tests, and they do not lag behind in the other tests, except for situations with insufficient memory bandwidth. However, NVIDIA solutions may be a little faster in games than in synthetic tests, as they do not require extraordinary computing performance.

Another conclusion, one of the key ones actually, is that we badly need DX11 synthetic tests. Preferably to replace synthetic DX10 tests from RightMark 2.0.

The next part of this article will deal with gaming tests. Performance in modern games must corroborate our synthetic tests conclusions. HD 5770 must perform on a par with HD 4870 and GTX 260, while HD 5750 should compete with GTS 250. Besides, we are going to see whether new cards' performance is limited by low memory bandwidth.

Write a comment below. No registration needed!