Anand's GeForce FX review

[Chalnoth]

I just have to say that I think Anand really screwed up on this one. I mean, he tried, he really did. He tried to pick the best apples-to-apples comparison he could, but I really think he failed miserably.

Here's the reason: The settings he chose pitted nVidia's GeForce FX with trilinear filtering against the Radeon 9700 using bilinear filtering. This is just bad reviewing. His image quality comparisons also appear to be very limited.

As a side note, though, I would really like to know the difference between "aggressive" and "balanced" anisotropic on the FX. With just a quick look on my GeForce4, there is no difference between the two.

[Hanners]

Originally posted by Chalnoth
His image quality comparisons also appear to be very limited.

Compared to Tom's Hardware's review and complete lack of screenshots, Anandtechs IQ comparisons were a stroke of genius.

I can't get to Anandtech right now, I guess it's being hammered, but where did they state that they were using bilinear filtering for the 9700s aniso? I must have missed that part.

[Hanners]

Originally posted by Hanners
I can't get to Anandtech right now, I guess it's being hammered, but where did they state that they were using bilinear filtering for the 9700s aniso? I must have missed that part.

Okay, I see this part now, and it does seem unusual, but to be fair Anand did compare all the different AF modes between the two cards on both an image quality and performance hit basis before he chose which to use for each card in the benchmark.

To quote Anand:
since there's virtually no perceivable difference in image quality we focus the majority of our benchmarks on NVIDIA's 8X Performance - Balanced AF vs. ATI's 8X Performance AF.

[digitalwanderer]

Hey, you have a point! :)

Unfortunately, I don't really think the bi/tri-linear screening will impact the test results much...but they should have been run tri to make it fairer.

Other than that I gotta say I really liked the AnanananananandaTech review the best so far. Very thorough, not sugar-coated, and pulling no punches.

[Chalnoth]

Oh, and that brings me to another problem I have with the review. I'm willing to bet that those FSAA screenshots are not showing what is displayed on the screen. Remember that with the GeForce4, the Quincunx and 4x9 modes are sometimes hard to capture. I see no reason to believe this isn't happening here.

And yes, I he did really try hard with the review. I just think he really screwed up in choosing to bench trilinear vs. bilinear.

[PreservedSwine]

Since this seems to be the thread for conspicuous accusations, how about this one?

Nvidia uses a 16bit Z bufer to get performance?NVIDIA, in contrast to ATI, does not plan for any setting options for the Z-Buffer depth in the driver, which means that the only way to work around this problem is when the application provides this option. This is the case with Serious Sam 2. With the console command /gap_iDepthBits=24, you can force a Z-Buffer that the GeForce4 Ti and GeForceFX can display without a problem. This is an easy way to get rid of the display problems in SS2.

It has been presumed that NVIDIA cards gain a performance advantage through the smaller Z-Buffer - this can be confirmed only in part. The benchmark results of the GeForceFX under 16-bit and 24-bit Z-Buffer reveal only a very slight difference. And there's still the disadvantage of the display error. By comparison, the GeForce4 Ti cards chalk up better scores with 16-bit Z-Buffer than with 24-bit.


And what's up with this??

Read the section "Adaptive Texture Filtering". The card changes the texture filtering while playing games just to improve performance (and reduce image quality). This would help benchmarks.

Seems very shaky from here:angel:

[Nv40]

I just have to say that I think Anand really screwed up on this one. I mean, he tried, he really did. He tried to pick the best apples-to-apples comparison he could, but I really think he failed miserably.

Here's the reason: The settings he chose pitted nVidia's GeForce FX with trilinear filtering against the Radeon 9700 using bilinear filtering. This is just bad reviewing. His image quality comparisons also appear to be very limited.




hehe

i thought i was the only one to notice this... WTF !
he benchmarked 8x aniso quality in NVidia vs ATi 8x performance
aniso (not quality) :rolleyes:

just because ->(he) didnt noticed a "diference" in both IQ..
he also told that there were no diference in IQ between
ATi AF 8x quality vs. Ati 8x AF performance. :rolleyes:

its a shame that noone posted screenshots at 30-45 angles
in AnisoF... it will be the best way to compare apple-vs apples

why anand didnt tried 16bit ATI vs 32bit Nvidia ? :rolleyes:
the diference will not be noticeable either... :angel2:

i think the best reviews were the HARDOCP and the extremetech.. the TOmshardware review was too short ,,
and it is an irony that they saw the GeforceFX as the new performance leader... hehe

im an nvidia fan ,but i think the overall winner is still ATI
with the radeon9700pro , at least for today games ,
in 3d aplications Nvidia should be the best one..

without AA+AF the Geforcefx clearly wins..
with 4xAA+8AF it (sometimes) score as good as the R300.
and other times (old games) its 10%-30% slower than
the radeon9700pro.. :o

also it would be interesting to see Doom3 benchmarks ,
i think the best scenario for Nvdia NV30 will be in heavy pixel shaders directx9 games.. (wich none exist yet).. :(

64bits and 128 bits bechmarks could be interesting too..
to see wich card is more future proof.. because as we know
the NV30 and the R300 are both direcxt9 cards , a none games
exist today that can be tested..

my message to Nvdia is... the NV30 is a good product ,but
its deseperately needs a 256bit bus!..... wihout the louder dustbuster... :)

fix that in the Nv35 ,and you will surely will have an r300 killer..
but i will not be surprised is the R400 ships with a 512bit bus!
so dont understimate ATI again . ;)

[Corben9]

Come on fact is now (not only judging from anands article ) that the NV30 isn´t what most hoped it to be ! so what ! Buy the best that suits u and for me thats gonna be A AIW9700Pro .For u it might be the Black and Deker fan equped desaster (judging from all the generated hype and hopes).

In anyway get over it the IQ in the 6 month old R300 is better then on the now half a year overdue GFFX. And the framrates with aa and AF show us that clearley! And even if it was bilinear on the R300 (althou I didn´t notice them saying that and it vouldn´t be necesarry coz the default on the R300 is triliner+Aniso!) fact is there is no visula differencne and if ATIs bilinar AF+AA looks better then nvidias Trilinar+AF+AA i stick with the one who offers me more for my money and this time it´s not nvidia! Not to mention that unlike nvidia Ati has a lot of video features to offer and HDTV output!

adioue Nvidia helo ATI
(well somehow I knew that vould happen - 3dfx anyone!:)

[saturnotaku]

Originally posted by Corben9
(well somehow I knew that vould happen - 3dfx anyone!:)

Could we please stop with the NVIDIA/3dfx comparisons? NVIDIA is not going to die.

A more accurate comparison, IMO is that the FX is to the 9700 what the Radeon 8500 was to the GeForce3. Better in some aspects, worse than in others. A card that was supposed to be a killer but wasn't. And look how ATI responded as a result. Hopefully NVIDIA can do the same thing now that the shoe is on the other foot.

[AGP64]

Since this seems to be the thread for conspicuous accusations, how about this one?

Nvidia uses a 16bit Z bufer to get performance?



How about this conspiracy theory from extremetech:

Again, we're left to wonder if it isn't a case of the driver team picking a select few games for which to optimize early on, and left the others until they're closer to their 1.0 driver release.

source: http://www.extremetech.com/article2/0,3973,846458,00.asp

or this one to back it up:

The UT2003 engine represents yet another favorite optimization target for hardware makers, and it would seem the GeForceFX driver team has spent some time ensure that this game performs well from the get-go.

And let's talk about the quality of the drivers...

experienced texture/polygon corruption in UT2003 and NFSHP 2. The texture corruption did not appear in UT2003 until you enabled 4xS, 6xS or 8xS modes. And when I did you can see the result above in the first picture. That pattern dances all around the screen as you move around no matter what map. Once I turned AA off the patterns went away.

source: http://www.hardocp.com/article.html?art=NDIxLDQ=

and from anandtech offcourse:

We followed the same procedure as we did with our AF investigation, and in doing so we managed to uncover a bug in the 42.63 drivers NVIDIA sent us for testing; it turns out that a bug in this build of the drivers causes visual artifacts if NVIDIA's 4XS Anti-Aliasing is enabled, and thus we had to exclude that setting from our tests.

source: http://www.anandtech.com/video/showdoc.html?i=1779&p=11

:angel:

[Pepsin]

About the AF comparison on Anand, I think it was slightly unfair. The GFFX was taking twice as many samples as the R9700Pro, providing that the FX actually DOES Trilinear aniso filtering in the Balanced setting...?

The Performance vs Quality mode on Radeon only selects between forcing Trilinear or Bilinear AF. It does not affect 45 degree angles or anything. So, since 8x bilinear = 4x Trilinear, they should have compared ATI's 16x Perf. to the setting they used on GFFX.

The difference between bilinear and trilinear aniso is extremely small, and you shouldn't only look at image sharpness to determine the quality, like Anand did. He should have checked for mipmap boundries. 8x Quality and 8x Performance produces an equally sharp image. So, I don't doubt that the images looked the same to Anandtech.

Do note that the performance difference between 8x Perf. and 16x Perf./8x Quality, is extremely small on the 9700 Pro, so it wouldn't have changed much, and, the 9700 Pro would have produced an even sharper image and looked better than the GFFX.

[Chalnoth]

Originally posted by Moose
Anand's review is the absolute best so far. (How anyone can do a review these days and not include IQ comparison pics is beyond me.) Nice job Anand!
Apparently you didn't pay attention very well. When benchmarking vs. ATI's quality anisotropic, the FX still wins in Anand's benches. The problem is that the main benchmarks were with ATI's performance aniso, which uses bilinear filtering.

And yes, thanks Spiritwalker for pointing out that he did bench quality anisotropic. Notice that the FX usually remains ahead until the fps drops below 60 (the "universal" playability point).

[Chalnoth]

Originally posted by Spiritwalker
If you want him to play that way then should he not have compared ATis 2xAA to nVidias 4x, because that would be a better reflection of the image quality in that case.

No, that's BS. The 2x modes of both cards are essentially identical.

[Hanners]

Originally posted by Chalnoth
No, that's BS. The 2x modes of both cards are essentially identical.

:confused:

With the Nvidia drivers used in all the reviews, the GeForce FX's 2xAA is virtually non-existent.

[gordy]

Originally posted by Chalnoth
No, that's BS. The 2x modes of both cards are essentially identical.

better look again.

[Spiritwalker]

Originally posted by Chalnoth
No, that's BS. The 2x modes of both cards are essentially identical.

You have got to be kidding right.
The 2x and Quincunx AA modes looked like they did nothing on the GFfx. And the ATi 2x compared favorably to the GFfx 4x mode.

I dont know what you were looking at.

[Spiritwalker]

here is a screen cap showing ATis 2x vs nVs 4x
http://members.shaw.ca/spiritwalker/main/anandaa.jpg

[Spiritwalker]

other than on the vertical surface the ATi version is actually BETTER

[volt]

Originally posted by Chalnoth
No, that's BS. The 2x modes of both cards are essentially identical.

And in which review was that ? Last time I checked it looked like it was never turned on. Why not give it up bud ? The "free" 2xAA sure is free, because there is no anti-aliasing at all.

[Demiex]

I want to know what happened to the Free AA modes up to 8X that Nvidia claimed. If that was true then the bandwidth restrictions wouldn't of been much of an issue, and i believe nv30 would've had the lead over R300 in AA benchmarks just as much as it did baseline.

Demiex

[PreservedSwine]

Originally posted by Demiex
I want to know what happened to the Free AA modes up to 8X that Nvidia claimed. If that was true then the bandwidth restrictions wouldn't of been much of an issue, and i believe nv30 would've had the lead over R300 in AA benchmarks just as much as it did baseline.

Demiex

You really need to know how to recognize marketing BS when you hear it. Nothing is free in 3D. Next thing you know, you'll be wondering when we can be expecting cinema quality gaming on the FX:angel2:

[ReDeeMeR]

Originally posted by Demiex
I want to know what happened to the Free AA modes up to 8X that Nvidia claimed. If that was true then the bandwidth restrictions wouldn't of been much of an issue, and i believe nv30 would've had the lead over R300 in AA benchmarks just as much as it did baseline.

Demiex


Ahem, those statements were bull**** from the start, Nvidia lied just to make ATi's product look worse.

[Demiex]

--------------------------------------
You really need to know how to recognize marketing BS when you hear it. Nothing is free in 3D. Next thing you know, you'll be wondering when we can be expecting cinema quality gaming on the FX
--------------------------------------
U've played Doom III alpha right? Thats pretty cinematic to me, and games will just get better. Have u seen EQ2 models? Wow...weren't they designed with GeforceFX in mind? Yep they were, but everyone would know that, seeing we all watch the launch replay.

Y would NVIDIA screw itself over like that, it knew that the benchmarks would immediate prove their remark about free AA false. They're not stupid, they're not poor, they had NO reason to claim that. I think its more feasable that NVIDIA had the design, just didn't implement it correctly. And me bein a software engineer in training, knows that a project that planned to take 3 months, could take a year to complete if u were to run into complications. They got scared by the 9700 and release it too early, again this not being their fault seeing how business's are dictated by board members who no nothing about development. That much R&D though, should've released a good product.

Demiex

[PreservedSwine]

would NVIDIA screw itself over like that, it knew that the benchmarks would immediate prove their remark about free AA false
Yeah, and politicians never lie either:rw:

Politicians and marketing dept's have much in common. They both rely on the public to have short memories....
Being "in training" you'll figure it out sooner or later, these guys are completley full of it:)

[StealthHawk]

Originally posted by PreservedSwine
Since this seems to be the thread for conspicuous accusations, how about this one?

Nvidia uses a 16bit Z bufer to get performance?

Tom benched SS2 with a 16bitZ and a 24bitZ, and the results were the same almost.

http://www.tomshardware.com/graphic/20030127/geforce_fx-12.html

how are the getting extra performance?