I think even the 8500 AF was a lot less demanding.

Yes, all R100 - RV280 AF implementation is very fast... but really ugly too.

R7000 - R9250

GF3, GF4 Ti and GF FX in HQ mode

Radeon 9500 - X850

Yea the DirectX 7/8 Radeons have fairly ugly AF. AF cannot even be performed with trilinear filtering. But it could still look nice compared to standard blurred mipmaps of other chips, and the speed hit was minimal.

honestly for the best AF you want GF6+ or Radeon x1k or better, it has edge independent AF and Transparent AA

candle_86 wrote:

honestly for the best AF you want GF6+ or Radeon x1k or better, it has edge independent AF and Transparent AA

Radeon X (X600, X800, etc) officially support transparency AA in later drivers, and it can be supported via registry hack on Radeon 9 series (although I've heard very mixed things about performance on the slower Radeons - I've never tried it on my 9550). For "the best" AF you want GeForce 8/Radeon HD 2000 series, as their implementations are generally great (GeForce 8 is technically somewhat better). Radeon X1000 series is generally a close second (it looks worse in D3D AF Tester than in real-life IME). GeForce 6 was nVidia's first part with RGMSAA support, which was an improvement over the FX (and previous) series, however Radeon X not only offers that, but temporal AA (and will do super-sampled modes in CrossFire) which (imho) look better.

However, few of these cards are supported in Windows 9x, fewer are universal AGP (or even available for AGP), and none of them support palletized textures (which again, isn't a huge deal, but may be worth considering depending on your specific needs). GeForce FX, by contrast, offers all of those things. Its "xS" AA modes also look pretty good, and its certainly an improvement over much older cards (e.g. Voodoo2, GeForce 2, etc) for image quality. Ultimately it comes down to what you need the machine to accomplish at the end of the day, as opposed to some unilateral "one size fits many" solution.

EDIT

Here's some AF sample patterns I took a while back:

Quadro FX 1700 (GeForce 8600-based):

GeForce FX 5800 Ultra:

Radeon X850XT:

You can also see more in the TechReport review of the HD 2900XT:
https://techreport.com/review/12458/amd-radeo … ics-processor/5

yea but Radeon R300 and R400 based GPU's dont support angel independent AF

I don't think I've noticed the filtering optimizations R300/R400 do. Geforce 5-7 are another story though. ATI was even frequently praised for their image quality compared to what NV was doing to compete. You can just set Geforce to HQ mode, but it has a speed cost.

NV 50 series drivers were the beginning of the infamous NV game and benchmark hack era. I'm not sure if HQ mode disables all of the cheating.

NFS porsche is one of my favorite games, seems work perfectly with the 9500PRO on 98SE

going back to AF, I gave it ago on all GPUs I have ready to go right now, drivers always with default settings,

GMA 4500 (win 7 with latest drivers)

Geforce 6150SE (Win 8 with forceware 307)

Radeon 9500PRO (win 98se catalyst 6.2)

Radeon HD 4670 (win 10 default driver)

Radeon HD 5850 (win 8.1 catalyst 15.5 beta)

I've never tried gaming with AF on the GMA 4500 but the rest looks fine during games,

Guys, what's with the number of petals? Why 9500PRO has 8, 4500 has 4 and 5850 has none? Is that the software implementation or the way hardware represents the picture? I'm confused. It would be really nice if anyone discribed how to read those diagrams.

RacoonRider wrote:

Guys, what's with the number of petals? Why 9500PRO has 8, 4500 has 4 and 5850 has none? Is that the software implementation or the way hardware represents the picture? I'm confused. It would be really nice if anyone discribed how to read those diagrams.

I think the 5850 result from the ones I posted is the closest to ideal, the other shapes are are a result of angle dependent optimization, so I guess the geforce 6 and 9500 PRO are trying to do it (performance optimizations) in similar ways, although I think the 6150 is better because it's a lot smoother

The petals are angle dependence, You can also see optimized trilinear filtering in the jaggyness of the mipmap transitions. The 6150 is essentially doing bilinear filtering.

With the DirectX 9 Radeons I think you can get better filtering with control panel tweaks. Not sure though. It doesn't really matter because it's quite hard to notice the quality reductions anyway.

candle_86 wrote:

yea but Radeon R300 and R400 based GPU's dont support angel independent AF

I don't think any card supports angel-independent AF (sorry I couldn't resist). 😊

True "angle-independent AF" didn't come about until later than any of the cards mentioned here (GeForce 8 and Radeon HD 5000 and up if you want to get picky; Radeon HD 2000-4000 are "good enough" though imho, and the X1000 series can get close enough via control panel options). Basically none of this is relevant for Windows 9x though.

brostenen wrote:

Ohh my... This escalated pretty quickly. :-D :-D I just wanted to know if I can get more muscle-power out of an GF4-TI4200 compa […]
Show full quote

Ohh my... This escalated pretty quickly. 😁 😁
I just wanted to know if I can get more muscle-power out of an GF4-TI4200 compared to my Radeon9600 on Win98SE. 😉

EDIT:
All I want to do, is to mount the card. Install the drivers and go gaming. What the choose, what to choose?
I do not want any FX5XXX line of cards. My FX5200 just don't run NFS-Porche-2K without any tweaking.
And tweaking is what I really don't want to do just to get one single game running.

If you already have it working with the Radeon 9600 then what's the problem? Or did I miss something earlier?

I agree with stermy - under 9x I'd rather have an nVidia card (be it GF2, 3, 4, FX) as the ATi drivers for 9x leave out some features that you would otherwise have in 2k/XP/Vista. I'm not at all familiar with NFS, so I've got no idea what is or isn't happening with the FX card there.

Stermy - I've done a few Intel IGPs too:

Intel GMA 950 on 965G

Intel HD3000 (iGPU from Core i5-2467M)

SPBHM wrote:

I think the 5850 result from the ones I posted is the closest to ideal, the other shapes are are a result of angle dependent optimization, so I guess the geforce 6 and 9500 PRO are trying to do it (performance optimizations) in similar ways, although I think the 6150 is better because it's a lot smoother

Technically yes the 5850 is ideal, but from in-game image comparisons the "square" that Radeon HD 2000-4000 will put up looks close enough that you probably won't notice without a microscope.

Examples from TechReport are here: https://techreport.com/review/12458/amd-radeo … ics-processor/5

To rip a few images for those who won't click:

D3DAF Tester:

HD 2900XT:

8800GTX:

X1950XTX:

In-game:

X1950XT "flat" (the other two will look basically the same)

X1950XT "angled"

HD 2900XT "angled"

8800GTX "angled"

The 2900XT is technically inferior to the 8800GTX as measured, but in practice they are both significant improvements over the X1950 and other older cards. GeForce 6/7 will/should behave more similarly to the X1950 in its "default" mode, cheating issues aside (see below link).

Also worth noting, the 2900XT's AF mode is nearly identical to the "high quality" (e.g. "control panel tweaks") on the X1950; I do not recall if the X850 offers the same mode with the latest drivers or not.

TechReport also has a comparison with 9800XT, X800, 6800 Ultra (with and without cheats), and FX 5950: https://techreport.com/review/6672/ati-radeon … s-processors/17 (all in-game with colored miplevels).

SPBHM wrote:

I think the 5850 result from the ones I posted is the closest to ideal, the other shapes are are a result of angle dependent optimization, so I guess the geforce 6 and 9500 PRO are trying to do it (performance optimizations) in similar ways, although I think the 6150 is better because it's a lot smoother

HD 5000 series (and later) Radeons indeed have angle-independent AF - but there is a visible line betveen the AF stages (the first one is clearly visible in D3D AF tester - depends on the settings).
In some games (depends on the texture type) this error is visible and it is really anoying.

havli wrote:

HD 5000 series (and later) Radeons indeed have angle-independent AF - but there is a visible line betveen the AF stages (the first one is clearly visible in D3D AF tester - depends on the settings).

I cannot reproduce it on my Pitcairn card.

actually I see it now, enlarged via the forum- what is happening to images here?

Does anyone have a Geforce2 result? I swore that had aniso too though I could be mistaken.

leileilol wrote:

Does anyone have a Geforce2 result? I swore that had aniso too though I could be mistaken.

GeForce2 does support it but I believe it is limited to 2x or 4x. I can't seem to find a control panel image.

Putas:
your screenshot is correct - the first AF stage border is there. Try to compare it with this one http://abload.de/img/beznzvu-1odujb.png
AMD have solved most of the AF problems of the HD 5000's on next generation. Radeon HD 6000 series and all CGN based (HD 7000 / 200 series / 300 series) are much better in image quality. Difference in AF tester is not that big but for most games AF defects are almost non-existent.

leileilol:
Yes, GeForce2 (and also GF256 + GF4 MX) supports 2xAF.
http://i.imgur.com/9tqSCQD.jpg

I notice a lot of shimerring with the HD5000s, more than that it's difficult to say; but to be fair I've also seen to much shumerring with newer GPUs;

going back to the 9500PRO AF, from the options I can change it looks like the default settings (used on my screenshot) are the highest possible quality, I made some shots with different choices on the catalyst settings, and they show some interesting variations using lower modes if anyone have any interest I could upload, but I think that's enough flower pictures for today 🤣

Awesome thread, glad that someone split it and now it is its own thing. After saying that, allow me to continue lurking in silence once more 🤣 .