Poll about image quality

[andypski]

My first attempt at a poll - treat me gently...

Based on some discussions in another thread I wanted to try a quick poll of B3D people's views of two images from a recent review. I hope this is in the right forum.

The URLs for the images are (they're quite large):

http://www.ixbt.com/video2/images/asus-ati/r9800-38-halo3.jpg
http://www.ixbt.com/video2/images/asus-ati/fx5900-5214-halo3.jpg

[edit] Ah. It looks like there might be problems with the direct links above - if you're having problems then go to the main review page, go down to the section on Halo and look at the top two images from Example 3 (the ones in the hallway).[/edit]

And for the whole review :
http://www.digit-life.com/articles2/radeon/asus-ati.html#p1

If you answer that you feel there are other significant differences then it'll be interesting if you will say what you think they are. Please vote before looking at other people's replies so that you aren't influenced by them.

Thanks -
Andy.

 

[Bjorn]

Voted for other significant differences but don't know what's causing them. The floor, looks like texture aliasing (GF FX).

 

[PSarge]

That floor definately has texture aliasing on it on the GFFX.

All the light-coloured flecks in the middle distance... looks like a badly sampled bump-map to me.

 

[tEd]

There is texture aliasing on both but more on the gffx.

Either the gffx does less filtering or has a different mipmap LOD or both

 

[Rodéric]

There's a lot of differences, but since I don't know the game and level, I've no idea which one is supposed to look right ^^

 

[Simon F]

The aliasing unit is clearly not working properly on the 9800.


As for the lights, I've no idea if they are supposed to be there or not.

 

[darkblu]

• there's a tex layer (?) behind the chroma/alpha-keyed elements in the corridor which is completely missing on the gfx
• the aliasing on the floor on the gfx is very apparent. weird thing is that it does not seem like the base texture has the aliasing, but more like the bumpmap - as if it's point-sampled and/or with a slight negative lod bias.

 

[Xmas]

The floor looks like a point-sampled normal map. And it looks like slightly different gamma settings. The FX seems to have a tiny bit sharper textures on the walls and ceiling.

I'm not sure whether the lights are supposed to be on or off, maybe they are flickering. This doesn't look like a rendering error as they certainly don't use a "light off" texture for the lights and then blend a transparent light texture on top of it (which could be missing), but use two different textures for the different states.

 

[Sxotty]

I played that level on my 9800 (pro bios) and it looks nothing like that, so... I said too many compression artifacts. I mean it really doesn't I know because there actually isn't that much bump mapped stuff, and when I noticed the floor was I walked around in circles crouching to see how it looked.

 

[jvd]

My first attempt at a poll - treat me gently...

Based on some discussions in another thread I wanted to try a quick poll of B3D people's views of two images from a recent review. I hope this is in the right forum.

The URLs for the images are (they're quite large):

http://www.ixbt.com/video2/images/asus-ati/r9800-38-halo3.jpg
http://www.ixbt.com/video2/images/asus-ati/fx5900-5214-halo3.jpg

[edit] Ah. It looks like there might be problems with the direct links above - if you're having problems then go to the main review page, go down to the section on Halo and look at the top two images from Example 3 (the ones in the hallway).[/edit]

And for the whole review :
http://www.digit-life.com/articles2/radeon/asus-ati.html#p1

If you answer that you feel there are other significant differences then it'll be interesting if you will say what you think they are. Please vote before looking at other people's replies so that you aren't influenced by them.

Thanks -
Andy.

Halo is the worse game you can pick for this . It was designed for the the xbox so of course the nvidia cards already have a leg up on this . We need to find a game that was developed equaly for each card. Perhaps doom3 forcing standard path (and making sure nvidia isn't cheating) and also half life 2 (forcing full dx and no cheating ) will give us a great idea. Till then we have to make do with what we know .

 

[andypski]

Halo is the worse game you can pick for this . It was designed for the the xbox so of course the nvidia cards already have a leg up on this . We need to find a game that was developed equaly for each card. Perhaps doom3 forcing standard path (and making sure nvidia isn't cheating) and also half life 2 (forcing full dx and no cheating ) will give us a great idea. Till then we have to make do with what we know .

I hope you don't take this the wrong way, but I would really like to try to avoid discussing things like this within the boundaries of this poll as they're secondary issues to the ones I am looking at, and they also tend to introduce too many emotional side issues.

This also wasn't designed as a poll about which image is 'better', which is why I left out any 'A is better than B because...' options.

This is all about comparing two images rendered on two pieces of hardware, analysing them for differences and seeing firstly whether people feel that they see noticable differences or not, and secondly whether they feel they have enough understanding of what might be going on 'underneath the hood' to identify the causes of any differences that they do see. It's not about whether one piece of hardware necessarily has some unfair advantage on this particular application or not.

 

[RussSchultz]

Without both cards attempting to render the exact same input, its tough to make any decisions at all.

 

[demalion]

Not sure what is going on with the missing visual effects for the FX.

The texture details look pretty similar as far as I can tell, with differing lighting interactions changing the details highlighted on them. At least, outside of whatever is going on with the floor (which does remind me of the aliasing I remember from a bump mapped terrain in some strategy game demo from a while ago which had particularly astounding aliasing), and what appears to be compression artifacting in the medium/far ceiling.

 

[Joe DeFuria]

• there's a tex layer (?) behind the chroma/alpha-keyed elements in the corridor which is completely missing on the gfx
• the aliasing on the floor on the gfx is very apparent. weird thing is that it does not seem like the base texture has the aliasing, but more like the bumpmap - as if it's point-sampled and/or with a slight negative lod bias.

That was my assesment as well. (Not nearly as technically specific, but I'm guessing there's a missing texture layer on the FX, and that a light or bump-map LOD selection is not being properly done wrt the floor.)

I know AA is not being applied...is trilinear?

If application trilinear is supposed to be applied, the "aliasing" issue on the floor may be a result of the light / bump-maps not "matching up" properly with the base texture?

 

[Graphics_Krazy]

Bumpmap filtering appears to be of lower quality on the FX.
Could they be lowering the precision used for calculations (on texture maps)?

Their LOD is also altered in Det 5x.xx

How does this look on a Geforce 4 card?

 

[Hyp-X]

There's a dotted line trough the lights of the door on the R9800Pro that looks like one that definately shouldn't be there.

The floor is definately different - it should be viewed in motion to judge how much is that aliasing and how much is "detail".

 

[OpenGL guy]

They definitely aren't the same frame. Also, it's notable that the site compares nvidia to itself and concludes that "the problem is solved". Shouldn't they be comparing to another vendor or the ref rast to make that conclusion?

 

[Zvekan]

With missing lights it is very difficoult to spot other differences, at least it is for me. Except floor that look very bad on nVidias card and I think it looks simply horrible in motion.

andypski - can you give more info about reasons for image differences?

Zvekan

 

[Xmas]

The second Halo image exhibits similar artifacts on the water surface as on the floor in the third one.

But I don't believe the missing lights indicate a texture layer missing. It wouldn't make sense to blend a "lights on" layer onto the "lights off" layer, which is clearly visible.

 

[andypski]

andypski - can you give more info about reasons for image differences?
Zvekan

I could give my thoughts on the matter, but I started this poll because I was more interested in other people thoughts

Given that the only thing that was commented on in this image in the review was the differing lights I wanted to see if people looking at the images were really aware of other differences or not, and perhaps could get a handle on reasons behind such differences. It seems that people generally are aware of other, subtler differences than those mentioned in the review, whether they know reasons for them or not. Perhaps this indicates a need for deeper levels of analysis in reviews - after all, it seems that the differences are there for all of us to see, wouldn't it be nice to know why?

Anyway, given who I work for I'm not sure if it is appropriate for me to give my opinions about these images in this forum, which is why I wanted to try to avoid turning this into an 'A is better than B' poll in the first place.

- Andy.