G80 indepth IQ reiew

[Gstanfor]

Trilinear was never designed to hide mipmap boundaries on angled texels, AF was.

Anyway, I see we have now entered the "weasel words" part of the debate. You can't contribute anything useful, so you twist, turn and hawk your snake oil arguments instead.

 

[BFG10K]

Trilinear was never designed to hide mipmap boundaries, AF was.

But you told us the better the AF the more mip-maps it has so therefore it isn't hiding them.

But now you're saying AF is supposed to be hiding mip-maps and therefore have less of them?

So which is it? Is AF supposed to have more or less mip-maps?



I'm wondering, is my reasoning and questioning too complicated for you?

Would you like me to re-word my comments in a simpler form for you to understand them?

Or perhaps English isn't your first language?

Anyway, I see we have now entered the "weasel words" part of the debate. You can't contribute anything useful, so you twist, turn and hawk your snake oil arguments instead.

:roll:

 

[Gstanfor]

Mipmap B O U N D A R I E S...

 

[BFG10K]

I'm convinced you have some kind of comprehension impairment as there is no possible way you could be doing this on purpose.

Your posts demonstrate a total and utter lack of logic and reasoning and also appear to be constructed from random English sentences.

I'm not sure how I would go about attempting to reason with a brick.

 

[Gstanfor]

Perfect AF would not use mipmaps as you said. Sadly, video cards are not yet perfect and mipmaps get used. Proper AF using mipmaps is exactly as I described it above.

The colors we see on the AF tunnel show the mipmap in use on that particular section. If they show pure color, that means mipmaps are not being blended (hence AF filtering is not taking place). If they are blended color like you see with nv3x, that indicates that mapmap filtering is occurring.

the IHV's naturally try and stop as much mipmap filtering as possible since their resources are finite and fetching texels from mipmaps takes resources, hence optimizations.

As consumers, we want to see AF that blends mipmaps as much as possible, not as little as possible, until such time as mipmaps are no longer required (don't expect that to happen anytime soon).

I'm not going to explain any further. If you don't get that and what has gone before, you will never get it -- either that or you are being paid not to get it.

 

[Keysplayr]

Originally posted by: Gstanfor
Perfect AF would not use mipmaps as you said. Sadly, video cards are not yet perfect and mipmaps get used. Proper AF using mipmaps is exactly as I described it above.

The colors we see on the AF tunnel show the mipmap in use on that particular section. If they show pure color, that means mipmaps are not being blended (hence AF filtering is not taking place). If they are blended color like you see with nv3x, that indicates that mapmap filtering is occurring.

the IHV's naturally try and stop as much mipmap filtering as possible since their resources are finite and fetching texels from mipmaps takes resources, hence optimizations.

As consumers, we want to see AF that blends mipmaps as much as possible, not as little as possible, until such time as mipmaps are no longer required (don't expect that to happen anytime soon).

I'm not going to explain any further. If you don't get that and what has gone before, you will never get it -- either that or you are being paid not to get it.

I think BFG is getting it just fine. You seem to be the one with some very damaged ideas. I know you absolutely need NV30 to do at least one thing better than anything ATI has, but by your own links and examples, you actually hurt any case you thought you might have had. From this point on, it's just fun to watch what you're going to post next. At least you have entertainment value (somewhat).

 

[beggerking]

Originally posted by: BFG10K

ppl, this is actually true. NV3x does have the best AF dispite its lackluster performance.

This is false. Both the R5xx and G80 have better AF quality not to mention it's free, unlike the huge performance drop exprienced by the NV3x.

it is true. Angel-independent AF was actually in FX series and it wasn't "optimized" as in the x1000 series... which was stupid at the time because the card didn't have the power to push it.

it was also using full precision on all its shaders all the time, resulting in seriously bad performance.

wiki

also
"The good news is that G80 completely eliminates angle dependent anisotropic filtering. Finally we have a return to GeForce FX quality anisotropic filtering.... On the right, we see ATI's angle independent 16xAF degrade in quality to a point where different texture stages start bleeding into one another in undesirable ways"

anand

 

[josh6079]

Originally posted by: beggerking

Originally posted by: BFG10K

ppl, this is actually true. NV3x does have the best AF dispite its lackluster performance.

This is false. Both the R5xx and G80 have better AF quality not to mention it's free, unlike the huge performance drop exprienced by the NV3x.

it is true. Angel-independent AF was actually in FX series and it wasn't "optimized" as in the x1000 series... which was stupid at the time because the card didn't have the power to push it.

it was also using full precision on all its shaders all the time, resulting in seriously bad performance.

wiki

So it was angle independent, less detailed than the R5xx and G80, and basically unusable because of the huge performance hit the card took when using it.

How does that demonstrate that it has the best AF?

also
"[/b]The good news is that G80 completely eliminates angle dependent anisotropic filtering. Finally we have a return to GeForce FX quality anisotropic filtering[/b].... On the right, we see ATI's angle independent 16xAF degrade in quality to a point where different texture stages start bleeding into one another in undesirable ways"

anand

And? Just because they claimed they can now return to FX quality filtering doesn't mean that the FX quality was better than ATi's. All it means is that Nvidia is now back to the FX quality it once offered and then some. It's a blessing for them considering they took a step backwards in the AF IQ department with NV40.

The Anandtech link you provided further supports BFG10K's discussion about less color rings being representative of better filtering. Gstanfor was wrong in saying the NV30's AF was better because its AF pattern contained the yellow ring at it's highest setting whereas both the R5xx series and the G80 don't. If you look at the pictures Anand supplied, both the R5xx and the G80 reach FX quality at 8x AF: Click

(There's your yellow ring G, it must be better!!)

 

[Genx87]

Hey guys, personally I think my 8800 has fantastic color saturation, AA, and AF.
Honestly this thing is awesome!

 

[Cookie Monster]

The Gf3/4/FX arent angle independent in the true sense of the word "independent". So is the G80 and the R5x0.

Both Gf3/4/FX and R5x0 suffers from the 45 degree syndrome while the overall shape looks like a more rectangular shape (especially the HQ AF on ATi). G80 also isnt "perfect" as many have thought because if you look at the 16xAF pic, its more octagonal shaped, while enabling HQ makes it somewhat roundish octagonal shape. However, G80 is the best AF offered so far in the GPu industry and could actually be noted as being angle "dependent" by not by much.

 

[ShadowOfMyself]

Entertaining thread Reminds me of why I come to the video forum to have some fun...

Plus, I got to know exactly how AF works for the first time

 

[BFG10K]

Perfect AF would not use mipmaps as you said.

If you concede this point why do you have such difficulty understanding that the fewer mip-rings/colours the closer the card is to reaching that goal and therefore the better AF it has?

If they show pure color, that means mipmaps are not being blended (hence AF filtering is not taking place).

Not only is this statement false but it has nothing to do with what we are discussing. We are talking about # of rings and # of colours and your fictional claims of the more the better.

As consumers, we want to see AF that blends mipmaps as much as possible, not as little as possible,

No we don't - that's the exact opposite of what we want. We want the AF filter to use as few mip-maps as possible.

This is trivial to demonstrate by looking at the differences between 0x and 16x and seeing that at each increment the number of rings/colours reduces and the rainbow moves back further.

Again I'll post the link.

And look, at 8x your precious yellow ring is back so therefore according to your logic 8x is better than 16x? Likewise following your logic 0x is better than 16x since it has the yellow and green rings of your beloved NV30.

:roll:

I'm not going to explain any further.

One can only hope.

 

[BFG10K]

it is true. Angel-independent AF was actually in FX series and it wasn't "optimized" as in the x1000 series...

Err, have you been reading anything in this thread?

Based on the AF tester results the R5xx has better overall filtering to the NV3x. The G80 then beats the R5xx.

it was also using full precision on all its shaders all the time, resulting in seriously bad performance.

Not only is this wrong (nVidia did absolutely everything they could to get developers to use mixed mode/partial precision) but it's also irrelevant to AF and hence the discussion here.

 

[GundamSonicZeroX]

This logic vs. illogic war (BFG's on logic of course ) is very confusing. I can't understand where Gstanfor is getting half of his ideas from.

 

[josh6079]

I can't understand where Gstanfor is getting half of his ideas from.

Join the club.

 

[GundamSonicZeroX]

Originally posted by: josh6079

I can't understand where Gstanfor is getting half of his ideas from.

Join the club.

Thank you

 

[jim1976]

LOL is all I can say.. I was missing for a couple of days and I'd probably would have missed another priceless thread.. BFG let him be ffs

Originally posted by: BFG10K
And look, at 8x your precious yellow ring is back so therefore according to your logic 8x is better than 16x?

Actually I prefer 8xAF over 16xAF and I use 8xAF most of the times.. BUT for another reason..16xAF squares the circle too much for my tastes.. But let's stick to the topic..

Rys does very good job imho and I respect his opinion very much.. It's far from a biased personality

 

[GundamSonicZeroX]

I can't wait to see a comparison using games!

 

[AmdInside]

I used to feel like B3D was biased, especially after Dave Baumman joined B3D. As a 3dfx employee, he would constantly bash NVIDIA. His anger carried on after 3dfx shut down. I got the feeling that he blamed NVIDIA for him losing his job or something. He turned pro-ATI and the site felt like it went more towards the ATI side instead of being neutral. Good info but too much reading between the lines stuff. There are plenty of other websites out there that are more in the middle (yellow?) so I stopped visiting them. I do enjoy their forums though as they have very intelligent discussions in there.

 

[Gstanfor]

I actually never realizied Baumann was ex 3dfx - that explains a lot...

 

[Wreckage]

Originally posted by: Gstanfor
I actually never realizied Baumann was ex 3dfx - that explains a lot...

He now works in marketing for ATI
http://beyond3d.com/forum/showthread.php?t=31585

So yeah....take it all with a grain of salt.

 

[Gstanfor]

--------------------------------------------------------------------------------
Perfect AF would not use mipmaps as you said.
--------------------------------------------------------------------------------


If you concede this point why do you have such difficulty understanding that the fewer mip-rings/colours the closer the card is to reaching that goal and therefore the better AF it has?

Because you are comparing apples to oranges bfg10k. mipmapping works differently to no mipmapping at all and AF filtering using mipmaps is based on filtering the mipmaps to find the closest possible value to that required and then using that value.

The only way to get what you want is to use one mipmap level only (effectively not mipmapping), which totally destroys the texel caching architecture built into current GPU's and destroys rendering speed.

 

[Gstanfor]

Originally posted by: Wreckage

Originally posted by: Gstanfor
I actually never realizied Baumann was ex 3dfx - that explains a lot...

He now works in marketing for ATI
http://beyond3d.com/forum/showthread.php?t=31585

So yeah....take it all with a grain of salt.

A salt-mine worth of salt more like it...

 

[josh6079]

Because you are comparing apples to oranges bfg10k.

Three different AF algorithms at their highest settings is an apples to oranges comparison? All he's done is examine the NV30s, R5s, and G80s highest AF modes and determined that the NV30 has the worst out of all three. You and beggerking seem to be the only ones still in disillusionment, what a pair...

I'm still not quite sure what you're trying to argue Gstanfor. You were just plain wrong about the more colors in an AF test meaning better AF filtering, wrong about NV30's AF being superior to the R5's and the G80's, and now are trying to divert attention to a completely different B3D member who didn't even write the review being discussed.

In your opinion, why is the G80's AF better than the NV30's?

 

[Gstanfor]

No, Josh. BFG10K is trying to compare mipmapped texturing and filtering to nonmipmapped texturing and filtering. That's where the apples and oranges comment comes from.