Something wrong with 9700?

[legion88]

(...)

I have 3DMark of course, but I can´t get any screenshots out of it. But please, have a look at iXBT:s screenshots from Dragothic. Look at the house in the center of the picture compared to the houses farther away at the left of the picture. Look at the ground. Comparing aniso 0 to aniso 16Q gives you what?
My old Radeon does considerably better than this, even if I can´t prove it to you. Look at the little animated comparison between 9700 and 8500; it´s too small to really tell, but it seems the 8500 has the edge here.

(...)

Regarding the "aniso 0" shot at iXBT, the rooftop shows that half of it is blurry while the other half is sharp. This is how it is supposed to look.

3DMark2001 comes equipped with reference images--assuming you purchased your copy. Just look at them and see for yourself.

 

[OpenGL guy]

Regarding the "aniso 0" shot at iXBT, the rooftop shows that half of it is blurry while the other half is sharp. This is how it is supposed to look.

3DMark2001 comes equipped with reference images--assuming you purchased your copy. Just look at them and see for yourself.

Those "reference" images are rendered by the HW just as the images in the tests are. They are done separately in the image quality tests to make it easier to compare between different platforms as they use the same data every time.

 

[Doomtrooper]

Yes explain that reference image had to be rendered on some type of hardware...like this article he did about the Xor tests..

http://www.legion88.org/

 

[OpenGL guy]

Yes explain that reference image had to be rendered on some type of hardware...like this article he did about the Xor tests..

http://www.legion88.org/

No, that wasn't what I meant. What I meant was that the 4 images rendered at the end are rendered with the HW. The application can always select the reference rasterizer to generate it's reference images, but I don't know if they are doing this.

 

[KimB]

You know what I would like to see for image quality comparisons? A reference image that uses sort of an "infinite resolution" rasterizer, that actually calculates the exact coverage of each texel on each pixel in the image. It would certainly take a while to render, and generally wouldn't be practical for bump maps/pixel shader effects, but should be great for normal texturing, with the same for edge AA (calculate exact coverage of each texture that shares the same pixel...).

In this way, you could give a card a score with max quality enabled (FSAA + aniso) based on how close it gets to this image.

It might get even more interesting if the score was designed to be based on some sort of human visual algorithm, where things like regular patterns that are much more noticeable get higher scores (lower score=better).

Update: Just realized that this would look pretty cruddy for magnified textures. It would be most ideal for significantly-minified textures, and thus the test could be built around those...

 

[AzBat]

No, that wasn't what I meant. What I meant was that the 4 images rendered at the end are rendered with the HW. The application can always select the reference rasterizer to generate it's reference images, but I don't know if they are doing this.

When I was helping beta test 3DMark99 I made suggestions in the area of the Results Browser. The reference images and their viewer were just a few things they implemented after my suggestions. At that time they told me that they generated those reference images by using the reference rasterizer and then included those files with the 3DMark installation program. I suspect they still do that today. So I doubt they generate the reference images post-install.

Tommy McClain

 

[legion88]

Regarding the "aniso 0" shot at iXBT, the rooftop shows that half of it is blurry while the other half is sharp. This is how it is supposed to look.

3DMark2001 comes equipped with reference images--assuming you purchased your copy. Just look at them and see for yourself.

Those "reference" images are rendered by the HW just as the images in the tests are. They are done separately in the image quality tests to make it easier to compare between different platforms as they use the same data every time.

What does that have to do with anything? Rubank wanted to know whether that blurriness he sees with the Dragoth pics (the rooftop) on the Radeon 9700 was the fault of the software or the fault of the hardware. The point is that it is not the fault of the 9700. That bluriness is supposed to be there (assuming aniso is disabled).