Messing with LOD bias... and a question about AA and consume

[Reverend]

There are two reasons for posting this.

One is the fact that some developers have found it fit to use aggressive LOD Bias in their games. At most times, and to the uninformed, this would seemingly result in graphics that look to be "more detailed". At other times, and to the informed, this results in aliasing that overrides whatever gains aggressive LOD Bias may "appear" to mean. There is no "law" governing developers from limiting LOD Bias in their games. Hence the need for AA solutions to solve the resultant aliasing.

The other reason has to do with users tweaking LOD Bias when afforded the chance (like the hardware they own that supports LOD Bias tweaking). Many (if not most) of them will like agressive LOD Bias while overlooking the resultant aliasing.

The question :

What would it take for a piece of hardware to be able to overcome aggressive LOD bias, whether introduced by a software by default or by a users' tweaking of it? MO is that the super-duper XXX-tap aniso won't solve this... it will come down to some form of AA... or something else I can't think of right now.

You will all know that aggressive LOD Bias is a feature likely to be implemented by "regular folks" when afforded the chance. To many, its apparent "more detailed" graphics overrides any introduced artifacts. That is the state of the overall consumer mindset about "graphics in computer games".

Note that, personally, I would wish for as an aggressive a LOD Bias I can use with a piece of hardware that allows me to adjust as such... but I am still bothered by the texture aliasing and would, hence, wish for something that can let me use incredibly aggressive LOD bias (whether by default by a game, or by my tweaking the video card's drivers if this is available) while allowing not introducing aliasing.

This is something that came to my mind while messing with the LOD Bias of my GF4 Ti4600 while playing Jedi Knight 2.

[edit]There seems to be a limit to the number of characters in the topic header[/edit]

 

[Roger Kohli]

Any chance of some screen shots to show us "lurkers" what the visual effects of changing the LOD bias are. Some of us come here to get educated you know.

 

[Bambers]

Pushing the mip LOD bias up (or down ) will give a shimmering effect on some textures near mip boundries.

Supersampling AA may solve this depending on the gfx drivers.

When it was first enabled on my gf1 there were 2 options for each mode, standard and high quality. Standard left the mip boundries where they were, high quality moved them out to match the resolution being rendered into the buffer. eg. 800x600 with 4xAA in high quality mode would use mip level suitable for a resolution of 1600x1200.

With a jittered/rotated pattern SSAA you could get away with using mip LOD for a res of 3200x2400 but really you'd just be having the same effect as aniostropic filtering.

It's less of a performance hit to force a standard LOD with a tweak utility and use aniso.

 

[Doomtrooper]

There is actually a big discussion about LOD bias on Rage3D as there is also a pretty big peformance hit by increasing the LOD. IMO this is also why doing game reviews with different cards there needs to be more emphasis on having the same LOD bias settings (this may sound out to lunch but I will show why).

These shots were taken on a Geforce Ti4600 default Lod setting in Return To Castle Wolfenstein.

Geforce Ti4600 4XAA DEFAULT LOD BIAS NO ANISOTROPIC

RADEON 8500 4xAA DEFAULT LOD BIAS NO ANISOTROPIC

Geforce Ti4600 4XAA LOD of -1.5 with 64 TAP Anisotropic filtering

RADEON 8500 4xAA/ 16X or MAX ANISTROPIC

Geforce Ti 4600 Peformance hit with:

4XAA LOD BIAS +1.5 43 fps
4XAA LOD BIAS -1.5 34 fps


So when benchmarking a video card, if it has a default lower LOD then for sure it will be faster. I never played with the LOD bias before as the ATI drivers only allow OGL LOD bias and I always leave the slider set to sharp but after looking at this I may play around with the slider and see how much speed you can pick up.


EDIT: I hosted the shots for now to make it easier to see the difference in LOD.

 

[Hyp-X]

One of the games that uses agressive LOD bias is SS2.
The shimmering was very annoying (for me at least).

Actually using Aniso does not eliminate the shimmering. While it allows the usage of more detailed mipmaps the game's LOD bias should always go for to high LOD (higher in this case).
It pushes back the shimmering a little becouse there's no shimmering as long the largest level should be used, and Aniso allows this to be used at a larger distance (usually in practical game situations).

There is no way to reduce this shimmering on GF3 cards as MSAA changes nothing on textures (except Quincunx - but that's not the right way to reduce aliasing).

As of Radeon cards, their Aniso looks like it's just a LOD bias increase + trilinear disabled... (I guess it's not, but that's how it looks like.)

Btw, Doomtrooper what those "Image Hosted by tripod" pictures meant to prove?

 

[Doomtrooper]

In case you didn't NOTICE read four posts up..Roger Kohli asked for screen shots showing the difference in LOD bias

 

[Sabastian]

Hmm that is interesting. Nvidia benifits from higher benchmarks as a result of having a lower level of detail according to what Doomtrooper points out. The difference in those screens is quite obvious. I guess ATi is upholding their IQ reputation and nVidia is living up to theirs. Thanks for the pics Doomtrooper.

Sabastian

 

[demalion]

Doom, I think all he is pointing out is that in almost all the pictures you posted, it says "image hosted by tripod" instead of showing a game screen shot.

 

[ERP]

I think that Aniso is the right way to solve the texture aliasing problem, leaving LOD bias at "the correct" setting. I also think that this should be done at the application level, not on a slider, yeah I know this isn't a particularly "PC" attitude.

FWIW - I did some tests recently, on the relative costs of Aniso and LOD bias in our game running on NV2A.
Adjusting the LOD bias was actually more expensive than enabling Aniso.
My guess is that the NVidia texture cache is heavilly optimised for the 1:1 Texel/Pixel case.

 

[Sabastian]

Doom, I think all he is pointing out is that in almost all the pictures you posted, it says "image hosted by tripod" instead of showing a game screen shot.

I can see them fine? Although I have seen the "image hosted by tripod" before. Shame you can't see them.

Sabastian

 

[tEd]

what i peronally would like to see is if SSAA is used as with the voodoo5 for instance that the driver would automatically set the lod bias for the best image quality so if you were using 4rgss on the v5 the driver would set a higher lod. i don't know how it's solved on the radeon8500 but so you could use ssaa and improve image quality without always manually tweak lod bias for every possible combination especially with the radeon8500 where you have up to 6 or more aa options to choose from.

 

[Bambers]

I don't think the v5 drivers do increase the lod. ATIs SV doesnt atm either.

 

[tEd]

I don't think the v5 drivers do increase the lod. ATIs SV doesnt atm either

i know i had a v5 myself , thats why i said i would like to see such a solution because setting a different lod everytime when changing the aa setting was and still would be painful[/quote]

 

[Basic]

With a jittered/rotated pattern SSAA you could get away with using mip LOD for a res of 3200x2400 but really you'd just be having the same effect as aniostropic filtering.

I wouldn't agree with that. RGSS might be equivalent to OGSS at that supersampled resolution wrt edges, but not for textures. The LOD should actually be equivalent to what would be used with no AA at slightly less than 1600x1200.

And regarding Doomtroopers images; I also get the "Image Hosted by tripod" on image 1, 2 and 4.

 

[tkopp]

Open new windows, view the picture properties, and then cut-paste the URLs into your new windows for a comparison. The radeon shots do look more detailed.

 

[Sabastian]

Does a lower LOD effect AA performance? Just curious about this really.

Sabastian

 

[Doomtrooper]

Oops sorry guys just checked the thread from my work laptop and noticed the tripod thing. The home system must have the images cached as they show up fine there.
Just follow Tkopp instructions and they show up fine.

 

[Reverend]

Any chance of some screen shots to show us "lurkers" what the visual effects of changing the LOD bias are. Some of us come here to get educated you know.

I would've assumed LOD bias is a very well known and discussed subject . Anyway, I'm at work right now but in a few hours' time I'll try and provide screenshots of Jedi Knight 2 and how combinations of AA, aniso and LOD bias affects IQ together with comments on my observations of all.

 

[tobbe]

Isn't Doom's shots showing the difference between supersamling/multisampling (and implicit LOD differencies) rather than an explicit difference in default LOD?
IE the LOD would be the same if AA wasn't used?

 

[Reverend]

Aw shoot, now I've gone ahead and done it... while taking all the screenshots, a thought crept into my head to actually make an article out of this... and so I shall.