Image quality - a blast from the past
- Thread starter AzBat
- Start date
They seem to have decided anisotropic filtering and Truform were too confusing for the control panel. However, support for the latter two are availalbe in the nwn.ini file (though the Truform option still has the same bugs that plagued the initial release of the game 
).
In the case of anisotropic filtering, their default restricted camera angles make benefiting from anisotropic filtering less evident usually. Though, for these and other reasons, their video options dialogue could do with some major changes in behavior IMO.
).In the case of anisotropic filtering, their default restricted camera angles make benefiting from anisotropic filtering less evident usually. Though, for these and other reasons, their video options dialogue could do with some major changes in behavior IMO.
Speaking of IQ, I need some feedback on the below images. Are they good, fair, poor? If so please explain your reasoning. I won't tell you the settings just yet but I will tell you the benchmarks. This is on my GF3 Ti200. Just some general comments would help.
Alley Serious Sam
Alley2 Serious Sam
Alley3 Serious Sam
Benchmark on SeriousSam Alley... with the IQ you see is 45 fps.
Suburbs
Suburbs2
Benchmark on Suburbs with IQ you see is 40 fps.
QuakeIII
QuakeIII2
QuakeIII3
QuakeIII ver 1.31 demo with quality you see 50.5 fps.
What do you guys and gals think? These are all at 1154x864 so make sure your browser is not shrinking them.
Alley Serious Sam
Alley2 Serious Sam
Alley3 Serious Sam
Benchmark on SeriousSam Alley... with the IQ you see is 45 fps.
Suburbs
Suburbs2
Benchmark on Suburbs with IQ you see is 40 fps.
QuakeIII
QuakeIII2
QuakeIII3
QuakeIII ver 1.31 demo with quality you see 50.5 fps.
What do you guys and gals think? These are all at 1154x864 so make sure your browser is not shrinking them.
Each and every single one looks blurry to me (focusing on the ground) fairly near to the viewer (EDIT: except the one near the edge of the platform). It is worse than I'd expect for 8x on the GF or 16x on the R300 based on past images, and certainly worse than 16x on my 8500. Not that they are extremely blurry, just "too blurry" for my taste. I could see someone considering them "good enough" quite easily though.
What's going on with them? If those are 16x or 8x images I'd be a bit shocked.
EDIT: Oops, you said GF3, and I didn't really answer your question. Well, I think the GF 3 can do better than that (yes, I'm focusing on the anisotropic filtering), so you've compromised the setting for performance. If I had a GF3, I could live with those settings and framerates, and to answer your question more properly, I'd say they are "good" out of a scale that includes "very good" (I'd place them "just above" "fair" on the scale you provided).
What's going on with them? If those are 16x or 8x images I'd be a bit shocked.
EDIT: Oops, you said GF3, and I didn't really answer your question. Well, I think the GF 3 can do better than that (yes, I'm focusing on the anisotropic filtering), so you've compromised the setting for performance. If I had a GF3, I could live with those settings and framerates, and to answer your question more properly, I'd say they are "good" out of a scale that includes "very good" (I'd place them "just above" "fair" on the scale you provided).
More correctly, it shifts the error in the signal from lower to higher frequencies. The aliasing is still there but is less annoying to our visual system.Chalnoth said:
A better solution is to have the pseudo-random displacements "parameterised". The parameters can then include screen position and time and so you can get 100% repeatability yet eliminate pixel-to-pixel and frame-to-frame correlation
Chalnoth said:
To garantuee that two renders of the exact same geometry from the exact same camera angle will look exactly the same. If it would change with time you'd see noisy artefact jumping around at the edges.
Repeatability is btw needed for conformance with OpenGL.
For those who are interested they are all 16 bit images. The Serious Sam images where 4x AA with 8x AF. The QuakeIII images where 4x with 9-tap Gaussian and with 8x AF. Quake III was at max quality except for being 16bit. Serious Sam was virtually Max quality except for being at 1152x864x16.
Doomtrooper
Veteran
Tricky, Tricky , Tricky
I was too drunk last night to even hope to guess.
I was too drunk last night to even hope to guess.
Interesting is that I get a tremendous speed boost with a insignificant amount of IQ loss. For example in QuakeIII with 4x AA and 8x AF (tri) max settings at 1152x864x32 I get 40 FPS running the timedemo. By switching one setting (32bit to 16bit depth) my frame rate goes up to 60fps. A whooping 20 FPS increase with (at least to me) virtually no IQ loss. Who knows, maybe I am blind but it does play good and looks good. Just an option which I don't think to many people use on Nvidia cards.
On a side note, I was going to order a Tyan Tachyon G9700 Wednesday or Thrusday of this week (after payday) since they where going for $329. Now after a few days they are up to $369!. Plus my Trinitron monitor just bit the big one yesterday so I ended up buying a 19" KDS monitor at (I hate to say it) at Wal-Mart. Actually my new monitor is better then I expected. Plus I runined my XP1600 processor by munipulating the bridges once to many times so now I am waiting for a Tbred XP1700 for my new machine. I wonder if I will ever own a Radeon 9700 pro!! .
On a side note, I was going to order a Tyan Tachyon G9700 Wednesday or Thrusday of this week (after payday) since they where going for $329. Now after a few days they are up to $369!. Plus my Trinitron monitor just bit the big one yesterday so I ended up buying a 19" KDS monitor at (I hate to say it
) at Wal-Mart. Actually my new monitor is better then I expected. Plus I runined my XP1600 processor by munipulating the bridges once to many times so now I am waiting for a Tbred XP1700 for my new machine. I wonder if I will ever own a Radeon 9700 pro!! .Doomtrooper
Veteran
Humus said:
But if the number of samples is large enough to guarantee no terribly significant changes in intensity on a per-frame basis for any given pixel, and the framerate is high enough (say 60 fps), then I think that changing the sample pattern for each pixel each frame could be of benefit. I suppose there would need to be a statistical analysis to get a good idea of what parameters would lead to an acceptable image, but once those parameters are realized, it should improve the overall quality.
That is, our eyes generally average what we see over about 1/8th of a second, so at 60 fps, that's just under 8 frames averaged together in our eyes. But, since our eyes will not report the entire scene at once, you will want every set of eight frames to have nearly the same average (assuming you're looking at a static image). Once this is done satisfactorily, the quality of the AA technique will be effectively multiplied by the number of frames rendered in a 1/8th second period (roughly...that number will depend on viewing conditions, as well as who's looking).
As for consistency being required for OpenGL conformance, I can buy that. I would just hope that if somebody does put out a renderer like this (which I feel would be fundamentally superior), then that will be changed.
