Radeon 8500 Aniso vs Geforce 4 Aniso
- Thread starter Dave Baumann
- Start date
nice way to nitpick and yet totally miss the point that this HAS BEEN DONE.Chalnoth said:
ATI is not doing rip-mapping, as you can tell if you do what i said.
Edit: I see Bambers has already shown you the truth
(posted shots).If it was rip-mapping, the shots would look identical as they are taken along the diagonal of the texture.
Althornin said:
My argument wasn't so much as to whether it was RIP-mapping or not (I think it's safe to assume that ATI's not using additional texture memory to approximate aniso), but as to whether it had problems with 45% angled surfaces. In this case, we know that ATI's LOD is independent of texture orientation. My question is whether texture aliasing is independent of texture orientation.
Now, from those shots, I'm still having a very, very hard time telling. The texture is too low-contrast to show any noticeable aliasing at a distance, but I do have to say that the floor near the center of the image looks like it could be very aliased indeed.
Notice that I'm using somewhat vague language as aliasing is only obvious when in motion...but it certainly looks like aliasing to me. A higher-contrast texture would show it much better. If you can do it in UT, it would be even better, as I could reproduce the shots reasonably-well on a GeForce4.
Btw, one last thing. I don't have much money, and I'm not going to go out and get a Radeon 8500 just to test it out. That's the whole point of message boards like this, so that we can get a better idea of what the hardware is actually capable of without actually buying it ourselves...if everybody just bought it and tried it out, their conclusions would be obvious, and there'd be no need for places like this.
Doomtrooper
Veteran
Thats nice Chalnoth but this subject has been beaten to death, Here, Rage3D and Nvnews where if anyone mentions the speed difference between the 8500 and Geforce 4 in Anisotropic filtering you are quickly replying 'but ATI's filtering is inferior' without even seeing it
So you ARE basing your arguement on what you read and not what you know or see.
So you ARE basing your arguement on what you read and not what you know or see.
Absolutely. But I also know that the speed of light is a constant. I can't say that I've personally measured the speed of light, or that I've gone through all of the rigorous mathematics to prove it without a shadow of a doubt, but I know it because I've seen the evidence.
The same happens here. I see the evidence that Radeon 8500 owners have presented, and I come to a conclusion.
If you really want the fullness of my conclusion, I can provide it, but I don't think you, or anybody else, cares that much, so I shorten it as much as I can. I know that to say that ATI's anisotropic is inferior is not 100% correct, nor does it tie in 100% with what I believe or have seen. What it is a simplification of a much longer conclusion in which people would get lost.
Speaking of which, right about now I could probably start blathering on about any subject, and people might not notice that I ever strayed from the anisotropic argument. But, I think I'll just stop.
The same happens here. I see the evidence that Radeon 8500 owners have presented, and I come to a conclusion.
If you really want the fullness of my conclusion, I can provide it, but I don't think you, or anybody else, cares that much, so I shorten it as much as I can. I know that to say that ATI's anisotropic is inferior is not 100% correct, nor does it tie in 100% with what I believe or have seen. What it is a simplification of a much longer conclusion in which people would get lost.
Speaking of which, right about now I could probably start blathering on about any subject, and people might not notice that I ever strayed from the anisotropic argument. But, I think I'll just stop.
As with anything IQ is subjective but one thing that is impressive with the 8500 is texture detail. I for one MUST use their AA with their anisotrophy to help curb the titles of texture aliasing artefacts, from the mip-maps and the aggressive lod, imho. Over-all it offers a very clean texture in most cases when combined -- but not every single texture is perfect but the postives out-weigh the few negatives in my mind. The hit is small over-all with anisotrophy and a perfect choice for the AA they offer, imho. Any other way with this product.. you simply couldn't enjoy the anisotrophy feature. The feature is enjoyed and improved with driver up-dates and glad they did it this way and a choice for a consumer to consider.
Pr will offer pr talk.... and spin. I am just glad there is choices to consider.
Pr will offer pr talk.... and spin. I am just glad there is choices to consider.
Jerry Cornelius
Newcomer
NVIDIA chips calculate the mip levels to use per pixel, as opposed to just applying it per pixel, as seen by the circular mip levels in the color mip level shots. Maybe there is something hardwired in there that keeps them from using the appropriate mip level selection when performing Anis filtering?
I'll wager a guess and say that Matrox will show everyone what Anisotropic filtering looks like when done "right".
I saw someone use the term "approximate" describing ATIs method. I think this is an incorrect use of the word. If their sample patch stretches in one direction more than another, it isn't approximate, it IS anisotropic.
Logic says that there is bound to be more texture aliasing in the ATI card if they are only sampling from a single mip level per pixel.
I'll wager a guess and say that Matrox will show everyone what Anisotropic filtering looks like when done "right".
I saw someone use the term "approximate" describing ATIs method. I think this is an incorrect use of the word. If their sample patch stretches in one direction more than another, it isn't approximate, it IS anisotropic.
Logic says that there is bound to be more texture aliasing in the ATI card if they are only sampling from a single mip level per pixel.
The only thing is, nVidia's anisotropic is about as good as it gets, as far as the look of the technique goes. Its limitation is in its performance, and in the maximum degree of aniso available. I'd be highly surprised if anybody could develope a technique that looked significantly, if at all, better at the same max degree.
Of course, I'm hoping that all future high-end video cards offer robust anisotropic (no preference over any angle...) with a max degree of no less than 16.
Bilinear filtering does not necessitate more texture aliasing. What it does mean is that the MIP LOD had better be a little bit worse than with trilinear to prevent aliasing.
Jerry Cornelius
Newcomer
Judging from the screen shots I hope you're wrong. In theory you are since I don't beleive there is any set standard, so eventually Anisotropic filtering will include near perfect ray cast sampling, which goes beyond just the ratio of the axis to one another.
With Anis the sample patch is larger so my guess is that most implementations use a better LOD bias. I can tell you that my Kyro mip level boundaries are pushed back when I force Anis filtering in the drivers.
It's really time for some reference images with near exact texture sampling to test these cards. The only reason we have bi, tri and mip levels is because actually filtering the image accurately is a lost cause on available hardware. With 16/32 and 64 sample filtering methods, I'd really be interested in seeing the XOR (or whatever it is) against a reference image. There must be some software somewhere that textures to very exacting standards.
Jerry Cornelius said:
I meant in the context of the same max degree of anisotrophy. Naturally, increasing the degree of anisotrophy into the future will further improve image quality.
The whole reason behind using MIP maps, bilinear, trilinear, and anisotropic filtering is to get the best possible image quality with the least calculation. There's no real reason to take 50 samples of an image when a lower-level MIP map would have already pre-sampled most of those. Yes, it would be nice to see what the actual differences are, but I don't think they'd be noticeable with modern anisotropic in most circumstances (The natural conclusion is that at very high angles/large distances, you'd notice the most difference...which will be remedied in the future with higher degrees of anisotrophy).
Please, stop stating opinion as fact - I think there are enough people in this thread who have already said the contrary as far as they are concerned. It seems you need to open your eyes a little.
DaveBaumann said:
People usually compare the GeForce3 at 8-degree aniso to the Radeon 8500 at 16-degree aniso.
Who here thinks that, just regarding visual quality, the Geforce 3/4's 8-degree aniso looks worse than the Radeon 8500's 8-degree aniso?
And, perhaps more importantly, you have to consider how much texture aliasing there is in each situation, as you can always adjust the texture LOD of the GeForce card if you'd rather have a little bit of aliasing...
Until it's shown that it is possible to use the exact same degree of anisotropic as the GeForce 3/4 and actually have less texture aliasing, I won't believe that you can possibly have aniso that looks better than these cards *at the same max degree*
Still, the GeForce 3/4's implementation does leave a lot to be desired in terms of performance.
So, what the basis for the comment?
Ever considered that you could do the opposite for Radeon to reduce aliasing?
Unfortunately, since triangles are batched, allowing the number of texture samples to change for every polygon will have adverse effects on performance, since setup costs will be higher, more state will need to be shadowed, and not as many things can be hardwired into the chip (a *lot* of optimizations happen in the hardwired stuff -- allowing a variable number of loopbacks per triangle could do some *really* nasty stuff to the FIFO).
In terms of hardware design, it's a whole lot easier to just take the worst-case number of samples, and then compute a filter on the fly that ignores the unnecessary samples.
This is a case where the software algorithm is a lot easier to implement than the hardware one... especially on a highly parallelized, deeply pipelined streaming processor like a programmable GPU. Getting great performance, great image quality, and a great price isn't an option -- one of those has to give.
Doomtrooper
Veteran
Chalnoth said:
Actually ATI 'sometimes' does allow Lod adjustment but also consider this..on the Rage3D thread where the A Geforce 4 card owner was matching LOD through his tweak utility (which required a Lod of -1.3 to -1.5) he lost 15 fps
Yes the higher lod causes some aliasing but any game I've seen it (serious sam is one) adjusting the 'In game lod control slider' 99% of the time eliminates it.
My personal peference is clear textures, I hate BLUR on a 21' monitor.
Doomtrooper
Veteran
Yep the slider isn't working though with the current offical drivers (Try it Dave), even putting the lod bias settings manually in the registry doesn't do anything..
I think 3286 or maybe 6015 was the last driver that allowed the slider to work
I think 3286 or maybe 6015 was the last driver that allowed the slider to work
Doomtrooper said:..on the Rage3D thread where the A Geforce 4 card owner was matching LOD through his tweak utility (which required a Lod of -1.3 to -1.5) he lost 15 fps
I just have to say that the picture I posted earlier was at -1 LOD. I shudder to imagine the aliasing at -1.3 to -1.5...-1 just looked terrible to me.
Similar threads
