New NV40 rumours(Stoled/Quoted from Uttar)

[oqvist]

Originally posted by Viral
Have you seen benchmarks where they test AA and AF individually? last i remember, ATI's AA generally took a bigger performance hit when enabled. This was comparing R350 to NV35, so perhaps memory bandwidth was the sole reason for this.

If what you are saying is true, and NV3x takes a larger performance hit from 8xAF than R3xx, yet R3xx doesn't take a larger performance hit from 4xAA than nvidia, then wouldn't benchmarks show nvidia being equal/ahead/closer to ATI with AA/AF off, but further behind when both are turned on? it should, however this has not generally been the case in any reviews i've read.

But you can´t compare 4x aa from nVidia to 4x aa on ATI. 8x aa is really like 4x aa on ATI. And you have to agree that the NV3X series compares better in term of performance without AA and anisotropic than with aa and anisotropic. That is of course if equal amount of aa and anisotropic is being used. Sure you can take 4x aa and 8x AF and compare it but then the NV3X will suffer in image quality.

[The Baron]

ATI's AA and AF both take less of a performance hit than NVIDIA's, IIRC. However, NVIDIA's AF offers higher filtering (usually true in D3D, always true in OGL unless you're running 16x AF and standing in a cube) while ATI's AA almost always (like 99% of the time) provides smoother edges. One of the major exceptions to this rule is alpha textures--NVIDIA's mixed MS/SSAA modes can antialias alpha textures, whereas no ATI mode can.

[StealthHawk]

Originally posted by The Baron
ATI's AA and AF both take less of a performance hit than NVIDIA's, IIRC. However, NVIDIA's AF offers higher filtering (usually true in D3D, always true in OGL unless you're running 16x AF and standing in a cube) while ATI's AA almost always (like 99% of the time) provides smoother edges. One of the major exceptions to this rule is alpha textures--NVIDIA's mixed MS/SSAA modes can antialias alpha textures, whereas no ATI mode can.

I thought that NVIDIA's FSAA(especially on NV35) had a much lower hit than ATI's FSAA? Although that is somewhat misleading because ATI has higher quality FSAA than NVIDIA.

[StealthHawk]

Look, you said that the gf3 was optimized for speed. Using lower quality filters does not "optimize" anything. NVIDIA did not tell their board manufacturers "use Quality X 2d filters." It was entirely up to the board manufacturers. There was no spec.

I believe that NVIDIA got stricter with the gf4, so there was a spec. And they got still stricter with the gfFX.

I am not saying that (at least some) gf4's did not have sharper displays because they used better filters. But you're making it sound like the gf3 sacrificed IQ at the cost of speed or something, which is entirely untrue. The fact of the matter is that the gf3 didn't have worse output than the gf2 or the gf1 or the TNT, so the word "optimize" is not being used correctly here.

Originally posted by oqvist
Of course it´s subjective. Image quality is a very subjective matter.

Then why are you declaring NVIDIA's shortcuts(brilinear) to be a big deal and also saying that ATI's shortcuts have no bearing on image quality :confused:

And ATI do support higher level of AF. And to complain about ATI not offering full AF when nVidia don´t do full Trilinear... Well... The image quality doesn´t suffer from ATI not rendering off angles and it´s still better than nVidias anisotropic in real games even if they are really equal.

You're making this sound pretty objective when you later admitted it was subjective. You just said above that ATI's AF looks better at equal degree AF than NVIDIA's does! Period.

It seems ridiculous to point out NVIDIA's shortcomings and make a big stink about them but at the same time say that ATI's shortcomings don't matter. If it is your opinion that ATI IQ is better than NVIDIA IQ because that is your preference, then fine. But don't dress it up as an absolute when it is a matter of preference. If ATI did not have any off-angle problems then you could indeed say that ATI's IQ is superior without any problems, as the two implementations would be directly comparable.

[Viral]

Originally posted by StealthHawk
I thought that NVIDIA's FSAA(especially on NV35) had a much lower hit than ATI's FSAA? Although that is somewhat misleading because ATI has higher quality FSAA than NVIDIA.

That was my understanding too..

[The Baron]

Hm. I think 2x has a relatively neglegible performance hit on NV3x cards, but it does very VERY little to the image quality. 4x is a much bigger performance hit, and the mixed modes are just painful.

(then again, I had a UT2003 demo run at a staggering average 14.93FPS at 1600x1200 with 0xAA/8xAF--my Icetomb demo really is the Destroyer of Worlds)

[oqvist]

Originally posted by StealthHawk
Look, you said that the gf3 was optimized for speed. Using lower quality filters does not "optimize" anything. NVIDIA did not tell their board manufacturers "use Quality X 2d filters." It was entirely up to the board manufacturers. There was no spec.

Why do you think I ever said that filters have anything to do with speed??? It has to do with image quality since I don´t like blurry image in high res!!

nVidia has ALWAYS opted for speed over image quality. The one doesn´t exclude the other. Great image quality is useless if you ain´t got the power to use it but there wasn´t really until the 9700 PRO there was a card that really could do both extremely well.

And yes as I have said it´s the manufacturers to put in those crappy RFI filters. Since nVidia was very loose on it´s QA. I am no fanboy!!! I am not trying to proove that nVidia > ATI or vice versa!!

I believe that NVIDIA got stricter with the gf4, so there was a spec. And they got still stricter with the gfFX.

Yes this was because the higher IQ in 2D and higher res in the 8500 make this obvious. Competition is great it really helps the development of video cards. Imagine if ATI wouldn´t have made their 9700 and the NV30 was the king of the hill. It would definiatly has been regarded as another success for nVidia since it is much better than a ti-4600. But then they would just continue using that core and the NV35 would be much more of a refresh of that one than it turned out to be

I am not saying that (at least some) gf4's did not have sharper displays because they used better filters. But you're making it sound like the gf3 sacrificed IQ at the cost of speed or something, which is entirely untrue. The fact of the matter is that the gf3 didn't have worse output than the gf2 or the gf1 or the TNT, so the word "optimize" is not being used correctly here.

Nope it didn´t have worse but if you read my post I said that nVidia has ALWAYS opted for speed over IQ or IQ through speed as you can always interpret it.

Then why are you declaring NVIDIA's shortcuts(brilinear) to be a big deal and also saying that ATI's shortcuts have no bearing on image quality :confused:

Because these optimizations are forced on you and do really compromise the IQ of the scene. ATI´s short cuts aren´t at all in expense of IQ in the same degree as the nVidia cards. This is also the reason why you see now magic 20 % increases in synthetic benchmarks like 3D Mark 2003 on ATI cards. And how the nVidia cards after the 3DMARK patch for some reason dropped remarkable in performance whereas ATI was quite untouched.

That just says it all. And yes looking at the tests you see where the speed improvements are being made.

The difference between optimization and cheating is that optimization makes your card run as efficient as possible whereas cheating you try to gain scores by sacrificing IQ.



You're making this sound pretty objective when you later admitted it was subjective. You just said above that ATI's AF looks better at equal degree AF than NVIDIA's does! Period.

Yup and it does IMHO!!! Period.

It seems ridiculous to point out NVIDIA's shortcomings and make a big stink about them but at the same time say that ATI's shortcomings don't matter. If it is your opinion that ATI IQ is better than NVIDIA IQ because that is your preference, then fine. But don't dress it up as an absolute when it is a matter of preference. If ATI did not have any off-angle problems then you could indeed say that ATI's IQ is superior without any problems, as the two implementations would be directly comparable.

ATI doesn´t got off-angle problems. That off-angle is one of the best features on ATI cards. Great performance boost at virtually no cost in image quality!!

And again. Explain to me how image quality can be all objective :confused:

If I write something I would assume everyone would understand that this is my opinion and not God almightys???

[The Baron]

Stealthy? Why do people not use endquote tags? It makes Mister B sad.

But I'll just answer this:

ATI doesn´t got off-angle problems. That off-angle is one of the best features on ATI cards. Great performance boost at virtually no cost in image quality!!

BWAAAAAAH?!

Go play a game with an organic environment. ATI's AF is very noticeably worse than NV's. Even HELLBINDER will admit that.

[-=DVS=-]

:fanboy:

Lets all go back to subject NV40 and R420 rumours :screwy:

[StealthHawk]

Originally posted by The Baron
Hm. I think 2x has a relatively neglegible performance hit on NV3x cards, but it does very VERY little to the image quality. 4x is a much bigger performance hit, and the mixed modes are just painful.

The performance hit of 4x FSAA got a lot smaller on NV35 over NV30, due to the great increase of bandwidth the NV35 has over the NV30. 2x FSAA has a very small hit on both, I think the hit is about the same, in fact.

[StealthHawk]

Because these optimizations are forced on you and do really compromise the IQ of the scene. ATI´s short cuts aren´t at all in expense of IQ in the same degree as the nVidia cards. This is also the reason why you see now magic 20 % increases in synthetic benchmarks like 3D Mark 2003 on ATI cards. And how the nVidia cards after the 3DMARK patch for some reason dropped remarkable in performance whereas ATI was quite untouched.

That just says it all. And yes looking at the tests you see where the speed improvements are being made.

The difference between optimization and cheating is that optimization makes your card run as efficient as possible whereas cheating you try to gain scores by sacrificing IQ.

Whether 3dmark is patched or not brilinear will be used. It is forced on for all D3D applications. Real trilinear is not possible because the driver is limiting the maximum IQ to brilinear...so your argument makes no sense as no 3dmark patch will enable trilinear.

Also, I don't see any "IQ compromises" in 3dmark03, but since you do, can you show me where they are? People have compared screens and no one has been able to find any significant changes since 44.03. And that, only when zooming in multiple times. I don't know about you, but those are not significant changes to me. With today's drivers, not even zooming can find any artifacting- so where are these "compromises?"

NVIDIA is cheating in 3dmark03 because changing the workload defeats the purpose of a synthetic benchmark. Not because of anything they did to IQ. With 44.03, IQ defects could only be found by zooming. Other IQ changes included equivalent results, such as trees in slightly different places. None of these could ever affect gaming. None of these will be visible to the naked eye. In that context, they can hardly be called compromises. After 44.03 even zooming yielded no IQ defects, but cheating was obviously going on.

I challenge you to show me the differences between brilinear and trilinear in UT2003. You will see that the brilinear used in Det50 along with application controlled filtering will have very, very minimal IQ differences compared to trilinear. NVIDIA should of course still offer the ability to use trilinear, but IMO brilinear is a good solution and the performance increase is well worth it. I think 99.99% will agree. The drawback is that it is being forced on people and that it is ridiculous that you can't use trilinear with a $500 card.

Originally posted by oqvist
ATI doesn´t got off-angle problems. That off-angle is one of the best features on ATI cards. Great performance boost at virtually no cost in image quality!!

No off-angle problems? That's ridiculous. Any angles besides 45, 0, and 90 will not get filtered beyond ~2x. In situations where such off-angles do not exist there is no performance gained as nothing is saved. Where performance is gained, IQ is lost.

And again. Explain to me how image quality can be all objective :confused:

Some quick examples: 1) Rendering of smoke. Card A renders it a certain shade of gray. Card B renders the smoke much blacker. The developer intended the smoke to be lighter. So Card A has the correct image, objectively it is correct, whether or not you actually prefer Card B's look.

Scenario 2) Full scene anti-aliasing. NVIDIA 2x FSAA vs ATI 2x FSAA. Discounting vertical and horizontal angles, both solutions offer equivalent IQ. It is a toss up between which is better. However, ATI clearly does a much better job at reducing edge aliasing on horizontal and vertical edges. As the purpose of FSAA is to reduce edge aliasing, clearly ATI's 2x FSAA is better than NVIDIA's because it does the job better overall. This is an objective decision.

If I write something I would assume everyone would understand that this is my opinion and not God almightys???

I try not to ever assume anything.


edit: anyway, DVS is right, this discussion is way off topic. So, I'll stop. All that needs to be done is to show proof that NVIDIA is making horrendous IQ tradeoffs and forcing them on people. Which may have been true at one time or another, but I don't think it's true now.

[fivefeet8]

Stealthhawk typed "Brilinear" so many times, it's funny. Isn't it "bi linear"? :D

[Paul]

Originally posted by fivefeet8
Stealthhawk typed "Brilinear" so many times, it's funny. Isn't it "bi linear"? :D

No, Brilinear is the phrase coined to describe nVidia's halfway solution between bilinear and trilinear.

[fivefeet8]

Originally posted by Paul
No, Brilinear is the phrase coined to describe nVidia's halfway solution between bilinear and trilinear.

Wow. It''s still very funny. Bri Linear? Ok. :eek: I always just called it the "psydo" tri-bi filtering. But brilinear is a lot easier to type. :D

[MUYA]

wasn't it 3Dcenter.de who coined the term "Brilinear" ?

[The Baron]

Originally posted by MUYA
wasn't it 3Dcenter.de who coined the term "Brilinear" ?
somebody at B3D, I think, and it spread like wildfire

[oqvist]

do you think the NV40 will support quadlinear :D

[StealthHawk]

Originally posted by The Baron
somebody at B3D, I think, and it spread like wildfire

Yeah, it was someone at B3D. God bless their wit :cool:

[Uttar]

Actually, both someone at 3DCenter.de (a reviewer) and someone at B3D (a forum poster, not sure who anymore) invented the term on their own, not knowing the other also did for a few days. The timeframe was very similar too! :eek:

Personally, even if I think it's someone at B3D who said it first, I'd rather give the credit to 3DCenter. At B3D, the person said something along the lines of "Tribilinear, Brilinear or BiTrilinear" (not sure at all about the other two, just inventing :p) while at 3DCenter, it was a full fledged investigation article using the term everywhere.

B3D invented it, and 3DCenter standardized the term, I guess. Before the 3DCenter article, pretty much everyone was using a different name for it, hehe :D


Uttar

[killermmn]

ha

[Soylent]

do you think the NV40 will support quadlinear

How about tri-cubic filtering? :P

[aapo]

Originally posted by StealthHawk
Yeah, it was someone at B3D. God bless their wit :cool:

I believe it was Dave H.

Originally posted by Dave H at Beyond3D
Still not clear on why "mixed bi/trilinear" doesn't work for ya'. In any case, both are too long for a permanent name for this (which we will eventually need). I suggest "brilinear". Or perhaps "lielinear".