My last post on this topic.
Crusher said:
What good does it tell you to know that one card is faster than another at rendering stencil shadows when calculating the geometry redundantly at intermediate steps in the rendering process, if no game is ever going to do that? Does it tell you how fast their vertex shaders are relative to one another? Or does it tell you how much each card stalls when throwing in redundant processing at different points? Or does it say something completely different?
If you read the release notes, Futuremark states that things were done this way to increase the workload. In other words, to
add complexity without having to create a lot more content.
It's easier to redraw something n times than to create n new objects. Does it really matter that it's the same geometry? No.
The one thing that you know it doesn't say is how fast Doom 3's vertex processing will be.
I
seriously doubt Doom 3 will be vertex limited in any sense. And guess what? GT2 and GT3
aren't vertex limited either! This has been covered on other web sites already.
So why would nvidia claim they
were vertex limited? My opinion is that it was a smoke screen.
Does it matter how well a card can render a single textured background with quad textured models if no game is ever going to use a single textured background with quad textured models?
Forgive me for being blunt, but this argument has no value whatsoever. What you're saying is that the application has to render everything
just like a game. Well, the end result may not
look like a game, but uses the same features. Many games
still have single textured pixels and the test represents that. In fact, the depth complexity of these single textured pixels is generally quite low, so I'd say it's pretty well balanced.
It seems to me like the only thing 3DMark03 is good for is telling you how well a video card is going to do at running 3DMark03. Whereas 3DMark2001SE seemingly had a little more relevance to the games of the past couple of years, and a much better chance at predicting how games released after it would be rendered, I see very little hope for 3DMark03 to be able to do this.
So tell us how 3D Mark 2001 SE did a better job. Also, give us the skinny on what future games will be like.
I guess the last point I would make is, if you really want a synthetic benchmark to compare the features of current video cards, then the benchmark should test each feature separately and completely independently of each other.
This is also a valid method for testing but no more valid than what 3D Mark 2003 uses. Interaction between states is very important as well.
By choosing game-like tests to compute the scores used to compare products, 3DMark03 inherintly shows they are focused on not showing feature comparisons, but gaming comparisons, and at that they fail miserably. If their scored tests for vertex and pixel shaders were more like their fillrate test, then I could see some legitmacy in the results. Instead, they've tried to fix the flaws that 3DMark2001 had (dependance on other parts of the system) in a day and age when the type of benchmarking methodology that 3DMark2001 used (make it similar to how they think games will be) is inadequate.
Who says it's not gamelike? nvidia? Please, do
your own analysis before making such claims.
3D Mark 2003 uses single texturing, multitexturing, stencil buffering, vertex shaders, pixel shaders and more. Since it's not a "gamelike" benchmark, I guess games don't use these features.
