[H]ardocp Review for the Geforce FX

[martrox]

Very nice review, Brent.

It's beginning to look like nVidia has really screwed the pooch on this one.

 

[demalion]

You tend to abuse the word objective, Chalnoth.

1. Angle problems. While improved significantly from the 8500, the Radeon 9700 Pro still has problems with off-angle surfaces.

The existence of these problems is objective. Your evaluation of them making image quality on the 9700 inferior to the GF FX is not.

But the 9700 is certainly inferior when comparing the same degree of anisotropic. If you want to talk about 16-degree vs. 8-degree, than I have stated many times that I felt that was subjective, but I side on the side of the GeForce4 (and presumably the FX).

Let me try this again, Chalnoth:

Even ignoring comparing 9700 8x to GF FX 8x, and focusing on comparing only 16x on the 9700 which you can atleast admit is "debatable", let's look at some example performance differences between 8x and 16x on the game that demonstrates the largest hit that I know of for the 9700: UT 2k3.

I don't think 84.18 compare to 87.34, 59.94 compared to 62.90, or 45.15 compare to 47.64 (as resolution increases) disqualifies comparing 16x ATI aniso to 8x GF FX aniso, do you?

Cutting to the chase past your "ironclade" 8x argument, we can clearly invalidate your conclusion of "absolutely better" concerning "In these situations, the FX's benefit will be in texture quality (high res + aniso will beat the texture quality of somewhat lower-res + aniso)", since nothing is stopping the user from using 16x aniso on the 9700 except your convoluted predetermination that the GF FX is better no matter what.

But back into the breach concerning your 8x to 8x comparison:

What is the criteria for "more accurate"? I'm guessing it has to do with the way you think colored mip map level gradients should look. So, where has the connection between this and your concept of "greater texture clarity/aliasing ratio" phrase been established, and then shown to be to such a degree that a user's evaluation is that the GF aniso is "absolutely better"?

No, not gradients, border shapes. The correct selection gives a sort of semi-circular pattern, which the GeForce series displays. The Radeon 9700 Pro does not. But, this is a relatively minor point, and will almost always not be noticeable.

Umm...I tend to agree. Though I'm not even sure the difference can be noticable with trilinear filtering, let's give you the benefit of the doubt and go with your statement: your admission seems to invalidate your use of it to support disqualifying the 9700 for comparison at high resolution, even with 8x aniso in comparison to the GF FX, don't you think? Let's not forget, however, that it has 16x aniso available.

I don't recall seeing an image where the 9700 didn't compare well to the GF in anisotropic filtering. Since the GF 4 image quality is "absolutely better", I'm sure you'll have no problem atleast providing sufficient examples demonstrating this clearly in contrast to those "rather pointless" images? Once you do that, we can cover the discussion about what "Balanced" aniso means, I guess.

Try looking at some shots in the Serious Sam: SE technology test sometime. Notice how the black brick borders on the ground look more jagged on the 9700 shots.

Umm...example? Perhaps I just missed it. If you say this is obvious and not open to another interpretation, and at the same time there are never advantages on the 9700 in comparison to the GF 4 "non adaptive" (using that as "GF FX balanced mode" equivalent even though we're not sure about that), perhaps you might have a case, but I don't see how you could based on shots I've seen. In my book, when different techniques have advantages in different circumstances, saying one is "absolutely better" requires a bit more than going with my personal favorite. Can you provide some indication of that "bit more" and preclude the existence of circumstances where the 9700 looks better even at the same level? Can you then go further and justify your selectively ignoring that 16x is legitimately comparable to 8x on the GF FX?

This is one example. More noticeable for me was in Neverwinter Nights on a similar texture. This effect is far more noticeable in motion than it is in screenshots.

And no such thing is ever noticeable on a GF 4?
Well, I don't have the cards to compare, so I guess your assurance that the GF FX balanced mode is identical to the GF 4, and that your comment about the 9700 is representative is what I have. I don't count that for much, but I'll leave the possibility that reality might conform to your bias open, and ask you again if you can preclude situations where the 9700 texture quality is superior to the GF 4, and whether you are disputing Xmas' comments about balanced for the GF FX being different?

But the 9700 looks better with AA, is the point.

Nice switch from "The FX's primary purpose will be to run at very high resolutions (1600x1200+) with lower FSAA, whereas the 9700 Pro would be better served by running at lower resolutions".

The AA is identical at 2x, and neither is playable at 4x FSAA at 1600x1200x32.

That was a good one Chalnoth.

First, let us ignore that some people might consider 1600x1200x32 with 4x AA and 8x aniso at 57 (performance) or 40 fps (quality) average a playable frame rate in UT 2k3 (or might actually not use the highest quality settings for physics and geometry). After all, no one could stand performance aniso even though UT 2k3 is full of situations with "low contrast textures".

Then, let's ignore the screenshots we have concerning the GF FX, and proclaim the quality identical.

Finally, we are left with the conclusion that the difference is to such a degree that the GF FX is the only choice. Hmm.

I can buy an argument that this is might be "disputable", but it is strange how when the comparison is something that would disadvantage the GF FX, ignoring screenshots seems to be the order of the day.

So it basically boils down to ignoring screenshots, proclaiming AA identical, ignoring the possibility of 16x aniso on the 9700, ignoring all games where 4x AA meets the undefined definition of "playable" on the 9700, and proclaiming the GF FX is the clear choice for 1600x1200x32 with 2x FSAA as a result? Just to make sure I properly understand you.

Hmm...so now games where you can play at high resolutions with FSAA are "pointless to consider"?

Yes, performance-wise, because it's going to be higher than it needs to be for good playability on either card.

The 9700 seems to tend to pull ahead with aniso and 4x FSAA at high resolution, and there are plenty of games where 4x AA + Aniso performance is rather high, and I was just confirming your evaluation that games where this is the case are "pointless", and possibly fishing for a justification besides that scenario disadvantaging the GF FX. Don't see it yet.

It only makes sense to consider games (in performance comparisons) that actually stress the video cards.

Hmm...if they didn't stress the video card (games that are "playable" under these circumstances), why does the 9700 sometimes pull ahead...is the GF FX turning up its nose in disdain at the games not worthy of its time?

On the one hand, the 9700 is delegated to be a low resolution high AA card when neither card is "playable" with high AA at high resolution (Let's presume all users share that evaluation of UT2k3 performance), yet at the same time games where you can play with high resolution and high AA don't count because their framerates are playable (so obviously, a user looking for a card to play games at high resolution would... never play them ). As a result, since a user would never use 16x aniso on the 9700 at high resolution, and there is no question that the balanced aniso mode on the GF FX is not changed in quality from the GF 4, and the 2x AA is identical between the cards (nevermind indications to the contrary thus far), the GF FX is the clear choice.

Well, and I thought your argument was biased.

Oh well, in all actuality driver changes might improve the GF FX situation and having assumed that as a given, even in the face of the evidence, might actually pan out for you. Good luck.

 

[Althornin]

Typo. Get over it.

Im supposed to know it was a typo?
Get over yourself.

 

[Mize]

Typo. Get over it.

Im supposed to know it was a typo?
Get over yourself.

Heh...yeah, I often hit the 16-key when I'm going for the 8-key...happens ALL the time!

Mize

 

[Novdid]

I'm new here, but let's face it...Chalnoth can't see anything bad in nvidia's products. All the posts I've seen from him indicates just that.

IMO after seeing those pictures and charts, the Radeon seems to be a superior product. The FSAA from the Radeon looks fantastic compared to the awfull results turned up by the GFFX. But then again the drivers that are driving the GFFX are rather raw so things can turn brighter for nv in the future.

 

[demalion]

While still technically inferior AA to the 9700, it seems established that the screenshots do not properly reflect the GF FX AA (except perhaps 4x OGSS modes if any of the shots are simply that).

Not that the possibility was discussed as part of Chalnoth's reasoning...but overlooking it was part of mine, though I granted Chalnoth's supposition of equality for the mode where quality might be closest, 2x AA, as part of my discussion.

 

[FrgMstr]

I agree with nggalai, Brent has brought a good injection of common sense into [H] in the reviews, little less PR spin..more screenshots, benchmarks that are important.

B3D vs. [H]

Lets get ready to rumble

You think me hiring Brent was an accident?

Brent showed lots of promise when we brought him on and he has certainly matured with his writing skills. His knowledge is very impressive compared to a year ago as well.

We have hired three mainboard editors of late, we will see who cut the mustard and be the next shining star.....

Also, we hired Sean "Pelly" Pelletier from NVNews as I saw some very good talent there that needed some direction and has great promise of turning into something very special. His last review covering the Shuttle box was one of the best I have ever read.


If the [H] is not changing and growing with the community, we will simply stagnate and become yesterday's news. I know for a fact that over the last year I have made considerable efforts to find local talent that has a passion for hardware and can relay that passion to our readers in the form of text on webpages. Trust me, that is something that is not easy to do (and certainly not easy to keep it going for five years), but we do our best and sincerely hoping we are getting better overall as time goes by.

 

[Doomtrooper]

The review quality has improved 100% Kyle, allowing full screen shots and the bandwidth that goes with that is great...
How not to do a review 101 is www.tomshardware.com

Gotta love those IQ comparison screen shots.

 

[WaltC]

While still technically inferior AA to the 9700, it seems established that the screenshots do not properly reflect the GF FX AA (except perhaps 4x OGSS modes if any of the shots are simply that).

Not that the possibility was discussed as part of Chalnoth's reasoning...but overlooking it was part of mine, though I granted Chalnoth's supposition of equality for the mode where quality might be closest, 2x AA, as part of my discussion.

I'm not at all satisified that what nVidia is doing at < 4xFSAA is actually FSAA. I don't think using the post filter qualifies. 3dfx, even with the V5, did not use the post filter to do 2xFSAA, and both the V3 and the V5 made use of the post filter (V3 moreso, because it was 16-bits only, and relied on post filter processing to approximate a faux 22-bit mode--which really did look better than normal 16-bits.) Not being able to catch it in the screenshots seems a dead giveaway to me (same as with the V3 initially.)

Apparently now we know courtesy of [H] that nVidia isn't going to talk about it. So I'm not at all sure that the screenshots not representing the screen in the case of an FSAA mode is at all any sort of redeeming characteristic.

 

[ClyssaN]

Brent is doing a great job

 

[noko]

I agree with Doom, the HardOCP was an excellent review and Tom's (the first one I read) was virtually worthless. With Brent's review you truely get a feel for what the card can do. It is written more objectively and matter of fact which I always like better. Just the facts and only the facts maam. I too am finding myself going to HardOCP more and more, for awhile I was just boycotting HardOCP, not that it really matters.

 

[Reverend]

So why didn't you do the review, Kyle? Isn't the "head" supposed to do such a review?

 

[Legion]

Chal, I can't quite understand your inner workings...

Due to drivers? That's utterly laughable, we all know drivers play a lesser part the higher in res you go. Time to stop making excuses for hardware shortcomings dude.

*G*

tell that to the PS2 developers.

Strangely, a friend of mine had problems such as these with a Geforce 2 GTS as it turned out it was a lack of power supply that was causing it. Could it just be this card is damaged?

 

[Nagorak]

Correct me if I am wrong but isnt the xS modes of AA only an option for D3D games? If so anybody that has run CS knows its runs so much better in OpenGL.

The fact that anyone is even talking about CS image quality is ridiculous. Anyone who has played CS (hmmm about 3+ years ago)... knows that the only thing that is going to improve image quality in that game is going outside to the tool shed, getting a sledgehammer, and putting it through your monitor screen.

First, let us ignore that some people might consider 1600x1200x32 with 4x AA and 8x aniso at 57 (performance) or 40 fps (quality) average a playable frame rate in UT 2k3 (or might actually not use the highest quality settings for physics and geometry). After all, no one could stand performance aniso even though UT 2k3 is full of situations with "low contrast textures".

No, the game is totally unplayable at those settings. That 57 fps is with nothing happening...in game you often will drop down into the 20s. Even the most performance-blind individual would not find the game playable at those settings. Even 1280*1024*32*4x*8x is essentially unplayable...very disappointing really, especially when everyone goes around proclaiming how you're not VPU limited until you have a P4 4.5 GHz. I guess my lowly Athlon XP 2100+ must be magical.

 

[Mulciber]

I dont really understand why unplayable framerates in a benchmark are anymore important than framerates in a theoretical benchmark. IMHO the two should be similarly classified.

 

[Ailuros]

...so I guess your assurance that the GF FX balanced mode is identical to the GF 4....

That's one thing I haven't seen so far up to 41.09 drivers and NV25.

Here's again the screenshot ram posted:

And here's the NV25 in "balanced" mode. I included the application settings:

By the way it looks exactly the same here with application and aggressive mode. Texture sharpening was disabled and driver LOD at "0".

 

[Ailuros]

I'm not at all satisified that what nVidia is doing at < 4xFSAA is actually FSAA. I don't think using the post filter qualifies. 3dfx, even with the V5, did not use the post filter to do 2xFSAA, and both the V3 and the V5 made use of the post filter (V3 moreso, because it was 16-bits only, and relied on post filter processing to approximate a faux 22-bit mode--which really did look better than normal 16-bits.) Not being able to catch it in the screenshots seems a dead giveaway to me (same as with the V3 initially.)

I'd say it's more related to that:

FSAA on GeForce 4 took a leaf out of the 3dfx technology books.

Normally with FSAA you have your FSAA buffer or buffers which contains information for all the pixel samples; so, for instance, if you are running 4XFSAA there will be 4 times the information in the frame buffer than will actually be displayed. When the frame is fully rendered its normal for each of the subsample to be read into the chip, averaged (to produce the final pixel colour from all the samples data), and then pushed out to the frame buffer again before being sent to the RAMDAC for display - this means that the FSAA sample buffer(s) is read into the chip then passed back out to another buffer that the size of the display resolution which is then sent back to the chip for the RAMDAC to display it.

What Voodoo 5 did, and now GeForce 4 is miss out the last step - it actually combines (averages) the FSAA sub sampled in the RAMDAC when the frame is rendered which means it misses on passing a buffer into and out of frame buffer RAM meaning less time is lost, but more importantly (in most cases) less bandwidth. It follows that Quincunx is also able to do the colour combination (averaging) in the RAMDAC as well which is why it produces results faster then the previous generation of GeForce.

http://www.beyond3d.com/forum/viewtopic.php?p=17490&highlight=quincunx#17490

 

[Crusher]

I was wondering if people could post some comparisons at an angle like this:

I'm interested in seeing what it would look like when applied to a ground texture, for instance.

 

[Ailuros]

It looks the same here at Level 2x. Is that a GF2?

 

[Crusher]

Yes, using the 40.72 drivers. I was wondering what the newer cards looked like from that angle, and with higher anisotropic settings. In particular, if the squareness of the outer mipmaps under Balanced and Aggressive settings on NVIDIA cards showed up much, or if the R300 looked any different (hard to tell from the quarter section of the tunnel view).