Jon Peddie Research - a technically oriented marketing, research, and management consulting firm in the graphics industry.

Testing Tessellation on the GeForce 480

Posted by By Robert Dow and Alex Garovi on May 28th 2010 | Comments Closed
Categories:
Tags:

Tessellation represents one of the key benefits of DirectX 11 for gamers. By enabling Tessellation during gameplay the GPU is able to “dynamically subdivide the wireframes of 3D objects.” By subdividing the wireframe the detail of all objects in the game is exponentially increased. Images that once took on a box-like look with tessellation become more naturally rounded. Tessellation can be programmed so that objects in the background that appear far away from the gameplay can be rendered with less detail while objects up close can take full advantage of the dynamic tessellation process providing max detail.

This week we took a look at Nvidia’s in-house benchmarks designed to measure DX11’s key ingredient for gamers. In the following Island benchmark developed by Nvidia the test is able to measure “Static” and “dynamic” tessellation rendering.

Now what does that mean? Static tessellation according to Nvidia is that the tessellation ration is equal for every polygon in the scene “This method does give equal amount of geometry detail for objects at any distance from camera.” This obviously will be more taxing on the GPU and really take no advantage of programming.

Dynamic tessellation calculates tessellation ratio depending on distance from polygon to camera. The closer the object is to the camera, the higher is tessellation ratio, details on objects that are near, and save GPU resources on objects that are far.

Nvidia developed two tests, Hair and Island. The test we used was Island. The test-bed was: 64-bit Windows 7, Corei7 3.3 GHZ, 3GB RAM, and a GF480.

We ran the test with various parameters on a GeForce 480 and an ATI Radeon HD 5970, and got the results shown in Figure 4.

From the testing we can see that Nvidia performed quite well. We would expect this considering it was test designed by Nvidia for Nvidia—either way the numbers are impressive. However we must keep in mind that the tessellation produced in these benchmarks far and away exceeds anything that would need to be rendered during optimal gameplay.

We asked ATI to comment on the results and we were told by a spokesperson who was obviously speaking off the cuff: “Any vendor can bake homegrown demos in such a way as to favor their products. That is exactly what is happening here. It is a demo Nvidia is seeding in their efforts to lead media and analysts to the conclusion that the 480 outperforms the 5870 in tessellation, with the follow on being that tessellation defines DX 11 (at least that is what they have been suggesting in some regions with a review program they are affectionately calling Real vs. Fake DX 11.

Yes, AMD’s view is harsh and we expect no less. We think the tests looked great, and “probably” create a situation that will not be found in a game for quite some time. So the value of comparing one AIB against the other with that test may not make a lot of sense for buyers looking for real world situations, but it does (nonetheless) show that Nvidia has a very powerful tessellation engine that can in fact handle outrageous models.