As an aside, I don't think I've ever seen trilinear filtering forced in hardware. It seems to me that it would be a fairly straightforward thing to do, and since there are a large number of games that do not have a trilinear filtering setting, it is something that I would like. As far as I can tell, it shouldn't be any different from forcing anisotropic filtering to enabled.
Anisotropic Options with 9700
- Thread starter Joe DeFuria
- Start date
Joe DeFuria
Legend
Rev,
Now I'm really confused.
I thought andypski explained it clearly myself...and I thought the app does NOT determine bilin or trilin samples unless "Application Preference" is used for anisotropic filtering modes....
Now, based on Dave's experience, what you are saying is correct. But IMO, Dave's experience does not jive with my interpretatoin of andypski's assertion.
So, andypski....what did you mean again?
Should the application be able to choose bilinear or trilinear samples when Quality Anisotropic method is selected? Or is trilinear suppossed to be "forced" in Quality mode?
Now I'm really confused.
I thought andypski explained it clearly myself...and I thought the app does NOT determine bilin or trilin samples unless "Application Preference" is used for anisotropic filtering modes....Now, based on Dave's experience, what you are saying is correct. But IMO, Dave's experience does not jive with my interpretatoin of andypski's assertion.
So, andypski....what did you mean again?
Should the application be able to choose bilinear or trilinear samples when Quality Anisotropic method is selected? Or is trilinear suppossed to be "forced" in Quality mode?
Joe DeFuria said:So, andypski....what did you mean again?
Why, just what I said of course, neither more nor less
More seriously, I've just had a chat with Dave to clear the interpretations up, so I'll let him update you guys... (Just to build the suspense
)As it turns out -
"You are both correct, and yet both wrong..."
Joe DeFuria
Legend
Well, at least it's good to know that I didn't ask a question with an obvious answer.
Bastard!
Bastard!
Yes, it appears that when anisotropic filtering is enabled, the GeForce4 operates as a 4x1 card (four pixel pipelines, one texture per pipe).
This is, by the way, one of the reasons I feel that the NV30 will use an 8x1 architecture...nVidia has stated that they are going to start focusing on better pixels, not more pixels. In other words, an 8x2 architecture would, most likely, only improve performance in non-aniso situations. Given the already rediculously-high fillrate, I see no reason to bother to optimize for non-aniso situations.
This is, by the way, one of the reasons I feel that the NV30 will use an 8x1 architecture...nVidia has stated that they are going to start focusing on better pixels, not more pixels. In other words, an 8x2 architecture would, most likely, only improve performance in non-aniso situations. Given the already rediculously-high fillrate, I see no reason to bother to optimize for non-aniso situations.
This isn't exactly what happens.
There are limits per clock on the number of texels that can be read into the cache along with the number of texels the pipelines can sample from the cache, when aniso is forced alond with multitexturing it is fairly likely that the later of these two constraints will be violated so it takes 2 or more clocks to read the texture sample.
This is one of the issues with forcing aniso in the driver (it forces the filter mode on all the texture stages), If the application enables aniso rather than it being forced in the driver then different sampling criteria can be selected where they make sense on different texture stages say aniso on stage 1 and bilinear on 2, aniso on 3 and bilinear on 4 at which point the chances of being able to read the texels in a single clock are much more likely and hence the perfomance hit is much lower.
http://www.beyond3d.com/reviews/ati/radeon9700pro/index.php?page=page16.inc
A little clarification on whats occuring in the present release of drivers.
A little clarification on whats occuring in the present release of drivers.
Thanks, Dave, that page is very informative on how the 9700 does anisotropic.
I find it very interesting that the MIP map selection algorithm is quite different when anisotropic is enabled. I am disappointed that the selection algorithm doesn't appear to have been improved over the Radeon 8500.
Anyway, notice how the MIP map boundaries are completely linear with anisotropic enabled?
By contrast, on my GeForce4, the boundaries are semi-circular or hyperbolic, depending on the orientation of the surface. It therefore stands to reason that it would be prudent to look for the most aliasing, when viewing a flat floor surface, either at the absolute center of the image, or at the far edges.
I find it very interesting that the MIP map selection algorithm is quite different when anisotropic is enabled. I am disappointed that the selection algorithm doesn't appear to have been improved over the Radeon 8500.
Anyway, notice how the MIP map boundaries are completely linear with anisotropic enabled?
By contrast, on my GeForce4, the boundaries are semi-circular or hyperbolic, depending on the orientation of the surface. It therefore stands to reason that it would be prudent to look for the most aliasing, when viewing a flat floor surface, either at the absolute center of the image, or at the far edges.
Joe DeFuria
Legend
Thanks for the update Dave...that cleared it up nicely.
Joe DeFuria
Legend
No, only in bilinear mode.
I thought Dave made it pretty clear that the mip-map boundaries are not linear with anisotropic PLUS trilinear enabled:
Come on Chalnoth...you need to dig deeper to find some 9700 shortcomings.
Chalnoth said:
Hrm - by my reckoning they are only linear with Bilinear. Trilinear always seem to be semi circular aniso or not, that screenshot may not be the best for it (the notmal one I would use outside wasn't showing the mipmaps colours for some reason).
My tests are at 640x480 on NV2A and they correlate pretty well with the above description (which is parafrased from the description I was given courtesy of someone who should know what's happening).
There may be additional work done during aniso that also causes fill reduction (say different cache read behaviour), or there maybe a minimum number of samples it takes. But alternating sample modes Aniso/bilinear on the texture stages will benchmark at WELL over 50% fillrate so it's certainly not just running like a 4x1 pipeline.
Chalnoth said:
I'll have to agree with Chalnoth. Compare the "No Aniso" and the "16x Quality Aniso" side by side; mipmaps are definitely curved without aniso, and linear with aniso. The lighting makes it look slightly curvy, but it is nowhere near as curved as similar distances of mipmap in the no aniso shot.
IIRC, the Radeon8500's mipmap boundaries were always linear, with or without aniso. It seems strange for Radeon9700's mipmaps to be curved without aniso but linear with it. Has anyone taken a picture of radeon9700's bilinear; is it curved or straight?
Similar threads
