question on nv30 and ati radeon 9700
- Thread starter nevidimka
- Start date
Huh? Single pass? Single Pipeline? Of course trilinear does not require multiple passes or multiple pipelines.nevidimka said:
You probably mean single cycle (or single TMU), don't you?
AFAIK all current NVidia and ATI products have TMUs each capable of outputting one trilinear filtered texel per cycle. So I don't see this changing in future products. Maybe they're even capable of adding some anisotropic filtering without a fillrate hit.
If I was properly informed, the Radeon 9700 is capable of processing four bilinear-filtered pixels per clock. So, it should not only be capable of doing one trilinear-filtered pixel per clock, but should be able to do one 2-degree trilinear aniso pixel per clock (or one 4-degree bilinear aniso pixel per clock).
There has been no information on the processing power of the NV30's pipelines, but I would expect it to be similar.
There has been no information on the processing power of the NV30's pipelines, but I would expect it to be similar.
hmm all graphic cards even the geforce 256 can do trilinear. but none of them even the gf4 ti can do them in a single pass single pipeline. they only attempt trilinear by using single pass double pipeline. and that is not a true trilinear. true trilinear is done in a single pass single pipeline..
thats y i've asked if the ati 9700 or the nv30 can do the true trilinear?
thats y i've asked if the ati 9700 or the nv30 can do the true trilinear?
OpenGL guy
Veteran
This doesn't make sense at all. Maybe you mean single cycle instead of single pass, but even that doesn't make your statement true. Trilinear filtering is simply a certain number of texels sampled from a texture. That's it. How many cycles it takes, or how many pipelines, doesn't make a difference at all in the result.nevidimka said:
well, some very wise guy told me about this once. i forgot about the details . but from what i can recall.. trilinears in todays graphic cards are done with a single pass but double pipelines. it have to do with a single pipeline not enough to carry all the values? for a trilinear. so double pipelines are used. so they arent true trilinear. trilinear when done in a single pipeline.. is much better IQ wise( more sharper ). the rampage was capable of this. it did trilinear in a single pipeline, single pass. 
nevidimka said:well, some very wise guy told me about this once. i forgot about the details . but from what i can recall.. trilinears in todays graphic cards are done with a single pass but double pipelines. it have to do with a single pipeline not enough to carry all the values? for a trilinear. so double pipelines are used. so they arent true trilinear. trilinear when done in a single pipeline.. is much better IQ wise( more sharper ). the rampage was capable of this. it did trilinear in a single pipeline, single pass.
The trilinear of every video card in existence today is done in "one pipeline."
Trilinear is, very simply, the weighted average between the filtering of two different MIP map levels. Some older video cards used such techniques as MIP map dithering to accomplish trilinear (such as the TNT), and I believe others used shortcuts to reduce memory bandwidth load (such as generating the lower-detail MIP map on-chip, not reading it from memory).
As far as plain trilinear filtering goes, there is nothing better than what we have in today's cards (except perhaps higher-precision pipelines, but I doubt those would make much difference at all for standard trilinear).
Some things that we may see in the future to further improve texture filtering quality include non-linear interpolation, such as bicubic.
Trilinear ONLY means to sample bilinearly from two adjacent mip maps.nevidimka said:well, some very wise guy told me about this once. i forgot about the details . but from what i can recall.. trilinears in todays graphic cards are done with a single pass but double pipelines. it have to do with a single pipeline not enough to carry all the values? for a trilinear. so double pipelines are used. so they arent true trilinear. trilinear when done in a single pipeline.. is much better IQ wise( more sharper ). the rampage was capable of this. it did trilinear in a single pipeline, single pass.
thats it
IF it takes 24358739593 passes, its still trilinear.
Who cares if it takes 2 pipelines? There is no MEASURABLE precision lost in combining the results!
ok u guys think i'm wrong... but the guy i leart it from got very technical. n what i'm saying is true to my knowledge. the true trilinear as stated by the guy.. is supose to make the image more better than todays trilinear.
i hope this works.. can u see the floor at the far end of the plasma blast.. how clear it is?? what feature makes it clear like that?? n thats not anisotropic. its a shot from the rampage card.
if u cant see the pic above.. follow this link
http://www.boomspeed.com/rashly3dfx/q31.jpg
i hope this works.. can u see the floor at the far end of the plasma blast.. how clear it is?? what feature makes it clear like that?? n thats not anisotropic. its a shot from the rampage card.
if u cant see the pic above.. follow this link
http://www.boomspeed.com/rashly3dfx/q31.jpg
Will the Radeon9700/NV30 pixel pipeline floating point ability allow for better blending of the texture samples when doing Bi/Tri-linear sampling? Also will this carry over to the anisotropic filtering as well?
Since these cards have higher pecision floating point ability will we see a difference?
Since these cards have higher pecision floating point ability will we see a difference?
I don't know exactly, but NV30 doesn't allow filtering (ie. automatic filtering, you can do it in the shader pipeline) of float textures, so I think the filtering itself is still only fixed point (but at least 12 bits per channel). However I'm still sure that you'll see a difference because of the higher precision.noko said:
Well this is my line of thinking:
The consequences as far as I can tell in having a higher precision texture sampling method will be less texture aliasing overall.
Now does anyone knows what precision is being used for the Radeon9700 compared to other GPUs/VPUs? Or is my reasoning flawed?
A bilinear takes four texture samples and calculates an average blend for one screen pixel.
A trilinear takes two bilinear samples, one in each mip-map and blends those using a algoritthm, 8 samples from the texture per pixel.
An anisotropic sample on the Radeon9700 can take up to 16 Tri-linear texture samples per pixel on the screen for an anisotropy of 16. Meaning up to 128 samples from the texture has to be taken and then blended using an algorithm or algorithms.
As the samples goes up so does the number of calculations. The more precision there is in this stage the less rounding errors will be inserted in the final texture color.
The consequences as far as I can tell in having a higher precision texture sampling method will be less texture aliasing overall.
Now does anyone knows what precision is being used for the Radeon9700 compared to other GPUs/VPUs? Or is my reasoning flawed?
Using higher precision in the filtering can reduce calculation errors. So if you have banding or noice that comes from lacking calculation precision in the filtering, then it's possible to remove it. (I don't think I've seen such errors though.)
It can't remove aliasing, since that needs a better filter pattern.
It can't remove aliasing, since that needs a better filter pattern.
How about pixel popping and texture aliasing? Not so much edge aliasing or the crawls. Any idea what precision is used for the filtering on today's GPUs?
This leads into my next question on using MSAA, what precision is used there as well for the samples taken?
Thanks for the answer, something that I don't think we have talked about before here nor do I think it is indicated in anyone's hardware specs either.
This leads into my next question on using MSAA, what precision is used there as well for the samples taken?
Thanks for the answer, something that I don't think we have talked about before here nor do I think it is indicated in anyone's hardware specs either.
