Paint.NET 101:
1) Download and install Paint.NET (http://www.getpaint.net)
2) Open image 1
3) Select Layers -> Add new layer
4) Open image 2
5) Double-click on Layer 2 to open the properties window
6) Select Blending Mode: Difference
7) Observe the differences.
Eg, I take the first two images:
Now you know where to look... See those white spots in the grass areas? They stick out like a sore thumb now. Clearly AI must have been fooling around with the texturing somehow.
Cool. I'm going to start playing my games with paint.net now.
Those images already appeared in a thread here - most people didn't see any difference.
So, if AMD and NVIDIA can increase performance and the difference is only seen by using something like paint.net, well they have my thumb of approval.
That's not the point I'm making here.
I clearly stated that I recommend that you turn these things on. In fact, I run Catalyst AI with the high settings even.
However, I noted that they did NOT turn them on for the nVidia cards in the benchmark.
So it's not an apples-to-apples comparison.
Either have optimizations enabled on both, or on neither... but not only on AMD.
That is only true if NVIDIA optimizations don't impact IQ visibly, which I'm not saying they do. I've never compared IQ with NV optimizations on vs off.
I'll go as far as saying nVidia's options impact IQ *less* than AMD.
Before the 5000-series from AMD, the IQ difference between NV and AMD was actually quite significant. AMD cheated the heck out of texture filtering. It was one of the reasons why the 4000-series remained reasonably competitive.
Now AMD cheats a bit less because they have better hardware. The maximum IQ is now also better than NV's, and pretty much identical to Direct3D reference. But it is not the default setting anymore (not since Fermi arrived).
While that might be true, it is also true that most reviewers had trouble to spot differences between AMD and NVIDIA IQ.
Prior to the Radeon 5000-series, we never even HAD a card that could do full angle-independent anisotropic filtering.
You are incorrect about the 5000 series doing *full* angle independent anisotropic- it may pass the angle test, it fails at fully sampling. nV fails at being fully angle independent, but fully samples- http://alienbabeltech.com/main/?p=12648&page=3
Also, the NV1X and NV2X parts both had full anisotropic filtering- although the performance hit was significantly larger then what we see today.
Look at the anisotropic tests when the 5870 was released:
Yes, it was bruteforce, so they didn't even care about any angle at all.
They're seperate options!No they're not:
http://en.inpai.com.cn/doc/enshowcont.asp?id=7626
As you can see, image quality is affected by enabling Catalyst AI. This can only occur when filtering and/or shader 'optimizations' (read: cheats) are applied.
Anisotropic/Trilinear Filtering Optimizations: If you want the fastest performance when using Anisotropic or Trilinear texture filtering in a game, you should enable both these options. This may result in a slight drop in image quality, however for most games this should not be noticeable, and in return you should gain some extra FPS.
The article you refer to seems to show that AMD uses some kind of mipmap LOD bias there (which would show up as a shift in the colour pattern in the other images.. but if anything, the 5870 images show the colours shifted more the OTHER way).
Did you have trouble comprehending my post? The 5000 series pass the angle test, which is what you just went through the trouble of posting- what it fails at is full sampling, which is what I linked to. You then post links showing that nV doesn't pass the angle test which is also the same thing I stated. AF is the single most important IQ aspect in every graphics card I have purchased this millenium; I follow it rather closely.
What is that even supposed to mean? If they didn't care about any angle it would be isotropic which by default means it wouldn't be anisotropic.
I'm not even going to bother to explain that.
There is your 'perfect' AF. Passes the angle test very nicely, doesn't it? The blurry textures is because AMD is undersampling- they are cheating in how many samples they are utilizing instead of cheating on degree of angle in terms of sampling as nV does.
I'd love to see you try. I'd love to see you explain how it is they are going to ignore all angles and compute the degree of anisotropy applied per pixel- I can safely assure you that such an accurate explenation would be major news in the graphics industry as a whole.
Looks to me like AMD just paints a different Moire pattern because they are less angle-dependent than nVidia is.
The level of anisotropy has more to do with partial derivatives than with angles. I'll leave the rest as an exercise to the reader.
Look outside of the red area for the 5000- there is no pattern displayed *at all*- just complete blur(gray).
You stated without using angles, I assure you- you will make big waves when you explain to the rest of the graphics world how to do it.
Are we looking at the same images?
I see a checkerboard pattern... which under certain circumstances may result in gray pixels
You see the same happening with the other two cards, but because of their angle-dependency, the moire pattern interferes with the gray areas, so it is not that apparent.
Firstly, that's not what I said.
Yes, it was bruteforce, so they didn't even care about any angle at all.
*A* pixel may end up gray if it is perfectly alligned using a bilinear filter or on a flat surface.
The surface for this test is not flat- that is how we know the AF is undersampling.
How are you going to factor your partial derivatives in a vacum that has no angles? If they don't care about angles at all, which were your words, how are they going to derive partial derivitives?
There are tons of shades of gray, depending on the alignment.
I'm afraid it doesn't work that way.
Also I'd love to hear you explain how to use an angle to derive a partial derivative...?
Your arrogance had me fooled for a while
the more and more you go on about this subject, the less convinced I am that you know what you're talking about.
Can the fanboys get back to getting their hands dirty polishing this new turd from nv ?
Can we get back on topic with the 465 being a steaming pile, that turns out to be a GTX 275 with DX11 ?
Honestly at this point, aren't we used to that with everything except the fastest parts this generation? The 5750, 5770 and 5830 looked very poor compared to the parts they replaced in the same price point. I'm rather adjusted to the mid range being far less of a value this generation then it was last generation.