Doomtrooper
Veteran
Radeon 8500 Aniso vs Geforce 4 Aniso
- Thread starter Dave Baumann
- Start date
Dave,
The colorize mip-levels are revealing. Looks like the LOD on the vertical surfaces are very close but on the Radeon8500 the horizontal surfaces show that the Radeon8500 LODBias is more negative giving the wider mipmap boundaries. This makes sense since the Radeon8500 can take more samples in the horizontal direction, up to 16 bilinear, where as the Ti4600 can only do 8, this allows the use of the more detailed mip-map level at a greater distance with the Radeon8500 which gives that more detailed look I would say
.
The colorize mip-levels are revealing. Looks like the LOD on the vertical surfaces are very close but on the Radeon8500 the horizontal surfaces show that the Radeon8500 LODBias is more negative giving the wider mipmap boundaries. This makes sense since the Radeon8500 can take more samples in the horizontal direction, up to 16 bilinear, where as the Ti4600 can only do 8, this allows the use of the more detailed mip-map level at a greater distance with the Radeon8500 which gives that more detailed look I would say
.
Look in the middle of the first mipmap transition, both seem to be in a very similar position; if higher LOD were used then this would be pushed back for one or the other AFAIK. It does appear that subsequent transitions are further apart on the Radeon, however its a little difficult to tell the relative starting positions becuase GF takes the curved boundry approach.
Dave,
Look again if you will on the walkway (horizontal) mip-levels. The first transition colorize Red on the Radeon8500 covers about the same area as the Ti4600 first and second transition (Red and Green). The Radeon8500 clearly has wider mip-map bands on the horizontal surfaces meaning a more negative Level Of Detail Bios. Which is appropriate due to the increase anisotropy ability of 16:1 of the Radeon8500 over the Ti4600.
The increase detail by the Radeon8500 on the building roof in the background anisotropically viewed can be explained now by the increase anisotropy capability of the Radeon8500 since both cards are using the same mip-map texture to render that polygon. This is one place where the 16:1 better filtering capability can be shown, this is not a LOD difference but a filtering difference.
Look again if you will on the walkway (horizontal) mip-levels. The first transition colorize Red on the Radeon8500 covers about the same area as the Ti4600 first and second transition (Red and Green). The Radeon8500 clearly has wider mip-map bands on the horizontal surfaces meaning a more negative Level Of Detail Bios. Which is appropriate due to the increase anisotropy ability of 16:1 of the Radeon8500 over the Ti4600.
The increase detail by the Radeon8500 on the building roof in the background anisotropically viewed can be explained now by the increase anisotropy capability of the Radeon8500 since both cards are using the same mip-map texture to render that polygon. This is one place where the 16:1 better filtering capability can be shown, this is not a LOD difference but a filtering difference.
Flipping between the Radeon8500 and Ti4600 bilinear samples I can note and identify more texture aliasing with the Radeon8500 in these still images. Still not convince this would be significant at 40FPS where the eye would tend to blend the fast differences from one image to the next, still it is there.
Noko,
I can see that, however what I was getting at is the point of the first transition (from non-coloured to red) on the horizontal plane (floor) both GF4 and Radeon are at very similar points (in the middle, they are different at the edge) – if you futz around an put negative LOD bias on through the drivers or something then that moves back.
Randell,
I’ve used Radeon 8500 for several months now and I can’t say that there has been anything distracting with the aniso filtering on it, nor can I say the times that I’ve used a GF4 (including now) has the filtering leapt out at me screaming – this level of detail is incrementally better or worse IMO, and I shall quote Democoder:
I think that’s absolutely correct. When you sit here an analyse the images to the n’th degree its easy to pick faults or highlight where its better than another – for instance its clear to me that the floor tiles outside of the centre of the image display more texture aliasing on the Radeon than it does on the GF4; have I noticed that before? No. So, has it been detrimental to my in-game experience? No – it clearly offers an overall increase in IQ over standard filtering, and what’s more is it bring it at performance levels that are not prohibitive for this card.
ATi have long since stated that their Anisotropic filtering implementation is ‘adaptive’ and Tony Tamasi is using this 10% figure to the determent of it, and saying that because they do everything there is better; however one has to wonder if it is the best thing to sample everything given the IQ and performance differences. Right now its up to the each individual to decide which they would prefer, however it going to be interesting to see how this evolves over time with different iterations or hardware.
I can see that, however what I was getting at is the point of the first transition (from non-coloured to red) on the horizontal plane (floor) both GF4 and Radeon are at very similar points (in the middle, they are different at the edge) – if you futz around an put negative LOD bias on through the drivers or something then that moves back.
Randell,
I’ve used Radeon 8500 for several months now and I can’t say that there has been anything distracting with the aniso filtering on it, nor can I say the times that I’ve used a GF4 (including now) has the filtering leapt out at me screaming – this level of detail is incrementally better or worse IMO, and I shall quote Democoder:
Democoder said:
I think that’s absolutely correct. When you sit here an analyse the images to the n’th degree its easy to pick faults or highlight where its better than another – for instance its clear to me that the floor tiles outside of the centre of the image display more texture aliasing on the Radeon than it does on the GF4; have I noticed that before? No. So, has it been detrimental to my in-game experience? No – it clearly offers an overall increase in IQ over standard filtering, and what’s more is it bring it at performance levels that are not prohibitive for this card.
ATi have long since stated that their Anisotropic filtering implementation is ‘adaptive’ and Tony Tamasi is using this 10% figure to the determent of it, and saying that because they do everything there is better; however one has to wonder if it is the best thing to sample everything given the IQ and performance differences. Right now its up to the each individual to decide which they would prefer, however it going to be interesting to see how this evolves over time with different iterations or hardware.
I understand what you are saying, the LOD adjustment sets where the first transition for the Mip-Map broundary starts in which is the same for both cards, the LOD Bios sets basically the width of each Mip-Map which is differenct. Plus I understand the differences in how ATI and Nvidia lays out the Mip-Map boundaries where as ATI does its rectangular method while Nvidia does a more spherical method.
Another consideration is that on the Ti4600 with max anisotropic filtering you maybe limited to 1024x768x32 resolution for good game play while on the Radeon8500 you could get the same FPS at a higher resolution such as 1152x864x32 or higher. Now with those factors in mind the Radeon8500 seems much more appealing to me.
In addition the Radeon8500 has another feature that helps textures and that is super-sampling. I wonder Dave if you could also show some screen shots between the Radeon8500 and Ti4600 with 2x AA with performance graphs. I hope you keep expanding this discussion by adding a little bit more in your comparisons. This is highly interesting, at least for me that is. Plus I think we all thank you for this interesting topic.
The 8500s aniso does have some texture aliasing but its only noticable on a few textures. On a paved/tiled surface the alaising doesnt show up at all.
The only real noticable aliasing is on textures like this one in Q3
http://www.jamesbambury.pwp.blueyonder.co.uk/anisoq3.jpg
warning for 56k users: 550kb file
The blue line is the first mip level but there are also ones that look like mip levels but don't match using r_colormiplevels and I assume these are where the number of samples taken is increased. Texture aliasing is most noticable in the 'corners' directly infront, although there is some at the sides behind the mip level.
Personally I don't really mind a bit of texture aliasing (I used to knock the LOD down a bit on my gf256) and when its only notciable on a few textures then I'd take the lesser hit than the perfect IQ.
For the FSAA. 2x performance greatly reduces the aliasing on the texture and 2x quality eliminates it. Obviously there is the performance hit from these though.
I like the way that ATi and nVidia promote full sampling (atis smoothvision and nvidias aniso) and selective sampling (nvidias MSAA and atis aniso) whenever it suits them.
The only real noticable aliasing is on textures like this one in Q3
http://www.jamesbambury.pwp.blueyonder.co.uk/anisoq3.jpg
warning for 56k users: 550kb file
The blue line is the first mip level but there are also ones that look like mip levels but don't match using r_colormiplevels and I assume these are where the number of samples taken is increased. Texture aliasing is most noticable in the 'corners' directly infront, although there is some at the sides behind the mip level.
Personally I don't really mind a bit of texture aliasing (I used to knock the LOD down a bit on my gf256) and when its only notciable on a few textures then I'd take the lesser hit than the perfect IQ.
For the FSAA. 2x performance greatly reduces the aliasing on the texture and 2x quality eliminates it. Obviously there is the performance hit from these though.
I like the way that ATi and nVidia promote full sampling (atis smoothvision and nvidias aniso) and selective sampling (nvidias MSAA and atis aniso) whenever it suits them.
Dave,
Thank you for that, I'm still not used to the level of texture shimmering with a Radeon8500 (even after AA & Aniso) after my V5. I just wondered if a Gf4 was markedly different. If LOD was adjustable on a Radeon I'd definetely make it less sharper a notch.
Thank you for that, I'm still not used to the level of texture shimmering with a Radeon8500 (even after AA & Aniso) after my V5. I just wondered if a Gf4 was markedly different. If LOD was adjustable on a Radeon I'd definetely make it less sharper a notch.
Randell said:
Texture shimmering was definitely a concern of mine when I moved from a V5 to a GF3. That said, I'm very happy with how well the GF3's aniso handles texture aliasing and I don't really miss the V5.
Randell said:AYe I know John
Sure I dont miss the V5 on terms of texture sharpness/speed etc, but I do see shimmering I'm sure I didnt used to see.
It depends more on how the game handles its textures than the API. A LOT of OpenGL games are Q3-based, so that probably explains the D3D vs. GL differences. And Morrowind, for example, doesn't show a lick of shimmering, though I am running it at 8x aniso. . .a bit slow, but looks awesome.
I have absolutely no idea how Tamasi came up with the "10%" figure (of screen display).
I'm more interested in what's right (or should be right) due to my previous comment about rolling in flight sims (or any other games where your angle view tilts left or right by 45% or more). There are many flight simmers, don't forget.
My opinion is that Wavey's "article" is unneccessary (hehe) - the main point should be what Tamasi means by "10% of screen display"... and then take it from there. I don't think many would argue with him saying NVIDIA is "doing it right".
And to the rest that says "the proof is in the pudding, it is the actual image quality that matters", again, my flight sim comment. I don't have a 8500 but I would imagine that I would be disturbed by the "on/off" apparent "aniso'ed" display whenever I roll my plane/craft/whatever.
Performance will matter of course but then everything is a trade-off depending on what's pleasing or annoying to every individual.
I'm more interested in what's right (or should be right) due to my previous comment about rolling in flight sims (or any other games where your angle view tilts left or right by 45% or more). There are many flight simmers, don't forget.
My opinion is that Wavey's "article" is unneccessary (hehe) - the main point should be what Tamasi means by "10% of screen display"... and then take it from there. I don't think many would argue with him saying NVIDIA is "doing it right".
And to the rest that says "the proof is in the pudding, it is the actual image quality that matters", again, my flight sim comment. I don't have a 8500 but I would imagine that I would be disturbed by the "on/off" apparent "aniso'ed" display whenever I roll my plane/craft/whatever.
Performance will matter of course but then everything is a trade-off depending on what's pleasing or annoying to every individual.
Bambers said:From never owning a voodoo card all the v5 shots that I've seen look blurry too me, especially those with 4xFSAA.
probably because you just saw those with either default LOD, which should be notched down to about -1.25 with 4xFSAA or even lots of infamous V5 shots with LOD racked up to +1.0 or so on certain sites to prove that the V5 is 'blorry'. I particularly remember some very suupect Giants:Citizen Kabuto shots.
Not withstanding the V5's lack or true Trilinear.
John,
Would low quality textures and 16 bit textures be more likely to exhibit aliaising than high quality 32 bit ones? My worst games for shimmering are DAoC on castle walls, Camelot everywhere etc and the latest EA F1 demo, though its pretty good at 4xQuality AA.
I'm not sure I agree with that. AFAIK the point of anistropic filtering is to increase texture clarity at acute angles - what the point of increasing the quantity of texture sampling where on non-acute textures i.e. ones that are just facing the viewport?
They may have improved the performance however I would doubt they have changed what they are doing significantly - otherwise I doubt Tony would have made the statement at all.
Similar threads
