View Single Post
Old 07-31-02, 02:34 PM   #12
Join Date: Jul 2002
Location: San Diego, CA
Posts: 505

1) Nvidia optimized -> Biased

2) "And you mean to tell me that a card that scores a 5,000 compared to a card that scores 11,000 means absolutely nothing?" It does mean simething, but it isn't easily quantifiable.

3) "Couldnt hope to resemble gaming scenarios?" Yes. Name three games that use point sprites, for example. And the gaming situations depicted are unrealistic.
It uses maxfx engine, right? Which means unrealistically built 'maps'... Weird polycounts and all. Not to mention that certain level geometries are unobtainable with the mapbuilding techniques that are used. Also, the textures are really lowres (on the environments). Another thing is that there are very few special maps (texture maps) used. No bumpmaps, no refraction maps, etc. Particles play a very little part as well. These are not situations seen in typical games.

4) "I guarantee you the person who scores 11,000 over the person who scores 5,000 will have a better framerate and be able to have higher detail in their, "real world" games. Please, prove me wrong."

Simple. Most feature tests are done exclusively. A pixelshader test has pretty much just the pixel shaders there. The Dot3 test has a handful of polygons, a lightsource or two, and the maps. The pointsprite test has only pointsprites. The vertexshader test uses only the vertex shader engine.

A card may perform EXCELLENTLY on these tests. But in real life scenarios, where any number of the elements of the tests are involved, the card may not do so well.

Lets say I have an Nvidia CorpDanXP 5000+ video card, compared with an ATI SpectralPower Extreme Maxx.

While the CorporalDanXP card, with 120003dmarks, totally owns the Spectralpower card, which has 80003dmarks in the individual tests, the SpectralPower card fares much better under conditions where the elements are mixed because of it's more balanced architecture.

Get the idea? Because the CorporalDanXP card does SO WELL in the individual (let's call them shader) tests, it gets a really high number of 3dmarks. The Spectralpower card, which perfoms fairly badly in the individual shader tests, gets a lower mark.

When entering a game where all the shader techs are mixed, however, the spectralpower card might own the corporaldanxp card.

So, yes. 3dmark is not a good benchmarking tool.
PCarr78 is offline   Reply With Quote