My R9700 experince and Aniso still "flawd"?
- Thread starter jb
- Start date
Althornin said:
That's complete bull. The deficiency of the 8500's aniso, and the lesser deficiency of the 9700's aniso are not performance optimizations. They are optimizations designed to reduce the number of transistors required to do the calculations. In particular, the 9700's aniso problems shouldn't affect performance much at all.
Gery said:
Do you own a TNT? I still have mine lying somewhere around here...
The MIP lines on the TNT are NOT identical to the poly edges. They look pretty much exactly like those I've seen in screenshots of the Radeon 8500.
I owned a TNT and Chalnoth you are wrong.
Proof coming soon.
Some proof:
Source: http://216.239.53.100/search?q=cach...per+polygon+mipmapping&hl=en&ie=UTF-8
edit:
NV3 = Riva128. d'oh.... still looking for proof.
Proof coming soon.
Some proof:
Source: http://216.239.53.100/search?q=cach...per+polygon+mipmapping&hl=en&ie=UTF-8
edit:
NV3 = Riva128. d'oh.... still looking for proof.
Mize.. no need unless you dont believe what JC has t say... (see above post)
Chalnoth, not knowing that, I lost all credibility in you.. I thought you were an NVIDIA encyclopedia...
edit:
I lost all credibility in me.. NV4..TNT apparently states per pixel mip mapping in its spec sheet. How embarrassing
TBC
edit ii
Finally some proof to support Chalnoth's statement.
Source: http://www.sharkyextreme.com/hardware/reviews/video/hercules_tnt2ultra/d.shtml
But then again they were advertising FSAA with the TNT2 Ultra
http://www.sharkyextreme.com/hardware/nvidia_ultratnt2_p/b.shtml
From personal experience and a bad memory I remember it being per polygon based, at least in D3D. Oh well.
Chalnoth, not knowing that, I lost all credibility in you.. I thought you were an NVIDIA encyclopedia...
edit:
I lost all credibility in me.. NV4..TNT apparently states per pixel mip mapping in its spec sheet. How embarrassing
TBC
edit ii
Finally some proof to support Chalnoth's statement.
Source: http://www.sharkyextreme.com/hardware/reviews/video/hercules_tnt2ultra/d.shtml
But then again they were advertising FSAA with the TNT2 Ultra
http://www.sharkyextreme.com/hardware/nvidia_ultratnt2_p/b.shtml
From personal experience and a bad memory I remember it being per polygon based, at least in D3D. Oh well.
I believe the Riva128 has per polygon mip mapping, but the TNT had per pixel.
http://archive.ncsa.uiuc.edu/Vis/Publications/SIGGRAPH98/PCChipsets.html
http://www.ritualistic.com/articles/reviews_stbv4400.shtml
http://www.pctechware.com/reviews/tntslot1-8.html
edit: whoops, never mind.
http://archive.ncsa.uiuc.edu/Vis/Publications/SIGGRAPH98/PCChipsets.html
http://www.ritualistic.com/articles/reviews_stbv4400.shtml
http://www.pctechware.com/reviews/tntslot1-8.html
edit: whoops, never mind.
Well, I very specifically remember seeing the MIP map lines when playing Unreal with my TNT. The algorithm was most definitely not "one MIP map per polygon." This was using Direct3D, and the MIP lines looked identical to Radeon screenshots I've seen, where when looking at a flat surface, it looks like two lines joining at the center of the screen.
I don't really know how this algorithm was accomplished, either. I do remember announcements about the GeForce's MIP map selection algorithm that labeled it as a "per-pixel algorithm." Toward that end, it seems probable that the TNT still sported a per-poly approach, but that doesn't mean it used one MIP map per poly (For example, it might have been done in the triangle setup engine, with the calculations done on a polygon level...but not necessarily one MIP map per poly...whereas the GeForce's calcs are done at the pixel level...).
I don't really know how this algorithm was accomplished, either. I do remember announcements about the GeForce's MIP map selection algorithm that labeled it as a "per-pixel algorithm." Toward that end, it seems probable that the TNT still sported a per-poly approach, but that doesn't mean it used one MIP map per poly (For example, it might have been done in the triangle setup engine, with the calculations done on a polygon level...but not necessarily one MIP map per poly...whereas the GeForce's calcs are done at the pixel level...).
OpenGL guy
Veteran
That's not how the Radeon 9700's mip line look.Chalnoth said:
Hellbinder
Banned
I dont give a **** if Nvidia has had bigfoot captured since 1977.. I am *Beyond tired* of this senseless attitude (by many people) that Nvidia does things *the right way* or the *only way*.. and everyones elses methods are *cheating* or *hacked* or *ammature* or any of the dozen ways I have heard it put.
It is even more *odd* being that i'd say 70-80% of the people around think that Ati has better IQ from a *gamers perspective*.
Also 8x Aniso on the 9700 looks *identical* to Nvidias 8x. 16x *clearly* looks better than Nvidias 8x method, especially in motion.
Is it just me, or does it appear to anyone else that the mipmap boundaries on the floor in the tilted shots are closer because the POV is closer to the floor in the tilted shots? I mean, is the room rotating around your POV, or, as it appears, a point somewhat above your POV?
I'd have to see the same shots on a card that supposedly doesn't have the anisotropy "problem" to know for sure that the change wasn't just due to the change in POV.
I'd have to see the same shots on a card that supposedly doesn't have the anisotropy "problem" to know for sure that the change wasn't just due to the change in POV.
Bigus Dickus
Regular
Once again, this could be settled easily by someone posting an identical screenshot from a GF4 for comparison. Matt was kind enough to do that for Nature when the 40.xx drivers were being questioned as to why that score went up so much (and the screenshots indicated it was not due to any IQ reduction).
If the GF4 shot looks the same, then the discussion is moot. If it's better, then we can proceed with asking why.
If the GF4 shot looks the same, then the discussion is moot. If it's better, then we can proceed with asking why.
Bigus Dickus
Regular
After looking at the shots again, it does look like the POV is closer to the floor in the rotated shot.antlers4 said:
GF4 SS anyone?
[edited to fix my quote /quote tags]
Hellbinder[CE said:
I called it an approximation. And it's not senseless. Approximations should, whenever possible, be done away with. nVidia has shown that it can be done better, and has been doing it better for three years (talking specifically about MIP lines/aniso, of course...note that the original GeForce supported 2-degree aniso). Why hasn't ATI done away with the approximation?
And, if you think for a moment, Hellbinder, I think you'll find a similar fault in GeForce cards. I'll leave that up to you (hint: FSAA is not it...).
Seems to me that there's no difference for the GF4.
Also that the R9700 changes from something that looks a lot better then 8X for the GF4 to something that looks pretty much the same , maybe a little bit blurrier but that might be because of the different angles in the screenshots.
Might add that i wouldn't consider this to be a big deal (= none at all).
Especially in the light of the current competition.
Also that the R9700 changes from something that looks a lot better then 8X for the GF4 to something that looks pretty much the same , maybe a little bit blurrier but that might be because of the different angles in the screenshots.
Might add that i wouldn't consider this to be a big deal (= none at all).
Especially in the light of the current competition.
While it is definitely better than what the Radeon 8500 offered, I still consider it a flaw.
In particular, in situations such as the one shown in Serious Sam, it's extremely noticeable. Flight sims would almost certainly show the issue more commonly (as has been repeated many, many times).
Now, if you want to attempt to make the argument that at its worst, the 9700 still isn't as bad as the GeForce4's max aniso, here's what I have to say:
1. The 45-degree angle is no longer the worst-case for the 9700. The worst-case is apparently at 22.5 degrees (according to previous posters on this thread), and every 45 degrees after that, which means to me that the 9700 shot posted at the beginning of this thread isn't necessarily a worst-case scenario.
2. While the approximation does make it so that the wort-case isn't as bad as the 8500's worst case, the flaw is apparently more common (I still consider it better than the 8500's, mind you...from sort of a "purity of rendering" standpoint...).
3. Visual quality is relative. Similar to how people have said they'd rather take 30fps constant over 20 fps minimum, 60 fps average any day. When those scenarios where the 9700 shows this flaw obviously come up, they will almost certainly be very obvious. I would go so far as to call the problem distracting.
In particular, in situations such as the one shown in Serious Sam, it's extremely noticeable. Flight sims would almost certainly show the issue more commonly (as has been repeated many, many times).
Now, if you want to attempt to make the argument that at its worst, the 9700 still isn't as bad as the GeForce4's max aniso, here's what I have to say:
1. The 45-degree angle is no longer the worst-case for the 9700. The worst-case is apparently at 22.5 degrees (according to previous posters on this thread), and every 45 degrees after that, which means to me that the 9700 shot posted at the beginning of this thread isn't necessarily a worst-case scenario.
2. While the approximation does make it so that the wort-case isn't as bad as the 8500's worst case, the flaw is apparently more common (I still consider it better than the 8500's, mind you...from sort of a "purity of rendering" standpoint...).
3. Visual quality is relative. Similar to how people have said they'd rather take 30fps constant over 20 fps minimum, 60 fps average any day. When those scenarios where the 9700 shows this flaw obviously come up, they will almost certainly be very obvious. I would go so far as to call the problem distracting.
I'd be glad to provide some GF4 aniso screenies in SSam : SE but it appears that aniso doesn't work (on my WinXP/40.41 system, at least) in SSam : SE thru either in-game-aniso-option or NV OGL control panel with my GF4Ti4600... any GF4Ti owners with the same drivers (40.41) having the same problem?
All I can say is this thread is going beyond nitpicking. I looked at both the R9700 and the comparison shots from the GF4 and I still can't see any noticeable difference between them at all. It I were forced to choose which looks better, I'd probably go with the 9700, but only because it seems to have better color.
I don't know what you people who are complaining are seeing, and it's my belief that whatever it is, it's not even there...
I could understand the complaints about R8500's anisotropic, although I didn't consider it a big deal. But this...is just completely ridiculous. The images are not totally identical, but the quality between them is essentially the same (and like I said above, if anything I think the R9700 looks better but not for anything having to do with filtering).
I'm sorry, this is just totally absurd.
I don't know what you people who are complaining are seeing, and it's my belief that whatever it is, it's not even there...
I could understand the complaints about R8500's anisotropic, although I didn't consider it a big deal. But this...is just completely ridiculous. The images are not totally identical, but the quality between them is essentially the same (and like I said above, if anything I think the R9700 looks better but not for anything having to do with filtering).
I'm sorry, this is just totally absurd.
I agree. I have to say to i have to see it in motion to be able to judge it i would consider it to be a problem though (f.e for flight sims, serious sam).
But i don't think that i would notice it in the majority, if not all games that i'm playing.
A question, does the same happen with the 8X setting ? (as in, reverting back to 4 or something under the special cases) ?
Similar threads
