The Best AA

[3dcgi]

The R9700 can AA lines - Parhelia can't.

Parhelia can AA lines so if it doesn't in a game it's likely a driver problem.
http://matrox.com/mga/products/parhelia512/technology/aa_vector.cfm

The R9700 can AA intersections - Parhelia can't.

True

The R9700 can AA with 24 bit Z buffer - Parhelia can't.

False. Where did this info come from?

The R9700 can AA with stencil buffer - Parhelia can't.

Half true. It can AA with a stencil buffer, but there is the possibility of artifacts so the effectiveness will depend on the game.

The R9700 can AA in windowed as well as fullscreen...

If so this might be a driver issue.

 

[Brent]

I voted R300's 6X AA


I haven't seen the parhelia in action so i can't comment on it

but i have seen the others and to me the R300 has the best AA IQ right now

 

[SirPauly]

I would say in some circumstances the V5's x4 has some advantages over x6 on the 9700, imho, but it is still early yet and ATI may tweak and improve the quality even more as time passes. The 9700 edges certainly make me forget about the V5 as well and to me they are quality edges over-all.

The V5 product offered incredible FSAA quality for its time and the strong points were never really discussed in an objective manner. All we did hear in many respects was camp talk and fluff reviews that helped fuel fluff talk and perception that many just did take as fact and never tried or bother to test the product first-hand.

If I had to hear RGSS FSAA was just a bit better than OGSS FSAA.... hehehe...... it can drive you bonkers in some respects when you had both products.

 

[no_way]

Which of the following gaming card* has surpassed it in terms of edge AA IQ?

right now, if someone were to offer me enough money, I would sell them my V5 6000. You see, not vapourware, you can actually buy it, of course, I'd only sell it for a good enough price, and what that is I don't know.

not vapour perhaps, but obviously your price will go even above professional workstation cards

 

[jb]

V5 x4 RGSS only because its the best method today that does AA all the time no matter what. The R9700 x6 is nice but any Alpha textures and jaggies. Same thing for the GF4. Matrox would have gotten my vote but there are cases when AA just is not applied. V5 was complete and had none of these issues. Once again 3DFX died but at least they left us on great thing, RGSS I know the V5 is too slow to play games today with all the eye candy, but for a tech that is well over 2 years old, it still does the job of AA better than any other method that we have seen. Note if R9700 did have a SS option (like i hear they are suppose to) then that would have been the one.. we shall see

 

[antlers]

I hope that non-OG SS does make it into the 9700 drivers. OGSS is the easiest kind to get right

 

[Randell]

well hopefully they can use the existing sample points from the MSAA and use them in an SS fashion or at least use the SS patterns from the original 8500 Smoothvision drivers, not the recent series.

 

[Ryu Connor]

The R9700 can AA lines - Parhelia can't.
The R9700 can AA intersections - Parhelia can't.
The R9700 can AA with 24 bit Z buffer - Parhelia can't.
The R9700 can AA with stencil buffer - Parhelia can't.
The R9700 can AA in windowed as well as fullscreen...

False
True
False
Application dependent
As can Parhelia

 

[easyride]

I have never seen any of the vote options in action

 

[Sharkfood]

STRICTLY speaking of edge-aa..

I'd have to go with the V5, but the R300 really does give a good run for the money given it's edge AA... generally speaking and not having the pleasure (yet) to put a Parhelia through it's paces. The poor Parhelia is probably going to be misrepresented in most polls simply due to the lower voting user base.

Obviously, once perception of edge-AA quality will have a lot to do with the software titles one plays. Games with lots of diagonal or opposing angle geometry versus lots of vertical and horizontal edges-only (ala most spots in first person shooters/quake3). NVIDIA's 4xOGMS/SS (GF3 &4/GTS->Ultra) and prior Radeons 4xOGSS still looks about as good as it can get on this kind geometry.

My litmus test mainly involves firing up Tribes (original) at 800x600 or 1024x768, fire up map Broadsides and run outside your base and view the enemy base from the safety of your own base's front ledge. The opposite base has three steps up the right side and edges of the gray base itself contrasted with the blue sky and green rolling hills. I still see edge crawling on every card to date, yet the V5 is near perfect.

PowerVR's Tubes demo is a pretty good test as well. Looking to clear all forms of edge noise/crawling as it moves the tube grid can show aliasing very well.

I have yet to see the Parhelia in action and would fathom it's 16x would likely be very nice in these conditions as well.

 

[DemoCoder]

Crawling existed on the V5 as well. Self tests are bogus because you know which card you are looking at and tend to concentrate more or notice more on one card vs the other.

The only real way to evaluate these things is to blind the experiment. Otherwise, if you know the card isn't a V5, you will "go looking" for the crawling, whereas you won't on the V5.


I also had a V5 and have a R300 PRO now and the R300 blows the V5 out of the water IQ wise, hands down -- WHEN PLAYING THE GAME. Playing at 1024x768 or 1280x1024 @ 6xFSAA and 16x aniso, I get an immediate "wow" over the V5. I don't have to slow down, look hard, or take screen shots.

It's the overall "look" that's simply better. Yes, if I devote brain cells during a fire fight to take notice of alpha textures, I can probably notice some problems, but I don't notice the 5% of pixels that are aliased when playing.

I do notice the lack of 16x anisotropic filtering and high resolution however.

Again tho, the best thing would be to run the V5 and R300 side by side and not tell the observer which is which and let them decide.


I bet 10 years from now, some small minority of folks are still gonna be claiming that the V5 has the best AA. Edge wise (-alpha texture issue) and opaque texture wise, there simply isn't any factual or theoretical basis for the V5 to stand on.

 

[Typedef Enum]

The R9700 can AA lines - Parhelia can't.
The R9700 can AA intersections - Parhelia can't.
The R9700 can AA with 24 bit Z buffer - Parhelia can't.
The R9700 can AA with stencil buffer - Parhelia can't.
The R9700 can AA in windowed as well as fullscreen...

Was this meant to be a joke? hehe

Since I'm probably one of the few guys to actually live/breathe w/ a Parhelia for any extended period of time, the issues with FAA are really not all that obvious...

Of course, we've been down that road a million times, analyzing screenshots to death, realizing how impossibly difficult it is to spot these issues when you're flying along @ 100 FPS...

OK, be that as it may...IMHO, the shortcomings of FAA are quickly rectified with the sharpness of the edges. And don't forget...this whole implementation still has a ways to go before more bugs are ironed out. I asked one of the Matrox guys some somewhat detailed questions about FAA and what's going on in the drivers...and as has been pointed out, they will tell you (OK, you have to sign an NDA) what those shortcomings are...In any event, they're still eliminating bugs from the implementation...and it's gradually getting better.

 

[Sharkfood]

A blind test would be the best way to determine which card has the best edge-AA. This would factor out any preference for one IHV, or even more appropriately, hatred for another.

Of course such a blind test would also be impossible to prepare as SS vs MS, the two test cases would be difficult to "hide" the differences in textures as well as LOD balancing. It's pretty hard to miss a supersampled scene versus one that simply has multisampling/edge-AA only. It would be difficult to balance out the two in such a way to provide similar texturing that wouldnt upstage the overall IQ one way or another.

Some people would focus on obtaining a perceived effect even if such didn't exist if they knew the card isn't of a brand that the subject dislikes.

 

[Clashman]

*cough* SCREENSHOTS PLEASE *cough*

 

[Randell]

I bet 10 years from now, some small minority of folks are still gonna be claiming that the V5 has the best AA

I hope not!

Edge wise (-alpha texture issue) and opaque texture wise, there simply isn't any factual or theoretical basis for the V5 to stand on.

You mean you the V5 doesnt have ethe best overall edge AA anymore at 4xRGSS? That overall the R300 6x is better, what about the R300's 4x?

From Types sceenshots I have to say the Parhelia FAA looks great. Matrox's next product, hopefully with a cheaper non-SG version may be very interesting.

 

[jb]

DemoCoder


the thread was about edge AA quailty so your point about AF on the R300 does not really apply here (although I agree with it).

Again I can not ignore the fact that as good as the R300 is, it does not do any AA on Alpha nor 16bit games. Try playing CS and you can see the see the jaggies on the alpha textures crawl quite easily which I find rather bothersome after awhile. Also there are still a few games that people play that are locked into 16bit color. While the rest of the world has move to 32bit color again until the drivers allow for this, the R300 does NO AA and that also should be taken in consideration.

Please dont get me wrong. I really like the R300 and its working more or less flawlessly in my system. But its AA which is the fastest and pretty complete still has a hole or two where as the V5 was "bullet proof" just too slow for most games today.

 

[DemoCoder]

V5's AA isn't bulletproof, especially at the resolutions you are forced to run it at, you can still notice some shimmer or flicker on edges. The V5 does not have "perfect" edge-AA, no way, no how.

Moreover, I'm not really concerned with edge-AA IQ on super old 16-bit games when every thing else is a joke (color artifacts, very low res textures, blurred detail everywhere)

In the case of counterstrike, the R300 does fine with the -32bpp command line argument.

The fact is, the R300 is fast enough on those old games to run 16X fsaa if the drivers allowed it. I can run 1600x1200 6xFSAA on counter-strike and still reach 60fps. ATI could put in a downsample pass to allow 6xFSAA multisampling followed by a 4X downsample to 800x600 for a grand total of 24x FSAA ( similar to NVidia's 4XS mode) and it would still run smooth.

I just don't think it's neccessary. IMHO, R300 6xFSAA surpasses V5. There are more possible colors for antialiased pixels, the sample pattern better hides bad angles, and the downsampling is gamma corrected. I can't think of any technical reason for how the V5's AA could possibly be better than what the R300 is capable of.


The alpha texture issue is somewhat of a red herring. Modern games are replacing alpha texture tricks with real geometry ( the fence or railing tricks) and atleast for me, the popular counterstrike maps that use it enough to notice are few (cs_italy)

Basically, the R300 has solved the aliasing problem for me, especially at high resolution. Maybe if you are running below 1280x1024 you'll notice it, I sure don't.

 

[Nappe1]

Since I'm probably one of the few guys to actually live/breathe w/ a Parhelia for any extended period of time, the issues with FAA are really not all that obvious...

as a sidenote, I am "jumping" back to the Matrox products after 2 years in ATI camp.

What was the problem?? general drivers? nope, games work like a charm. Product support?? well, a partly... I have two times tried to contact ATI support and twice only find out "ODBC cannot be contacted right now. try again later." another point have been these so called bonus features at my AIW and their drivers... but everyone have read that already in Saem's 9700 AIW Pro thread so, I don't think I need to talk about those here...

My tax repayment should arrive on December... Only thing I am right now wondering, if I will get a parhelia that is already in stores or maybe New Revision... I am looking forward to see, if Matrox has something new to show at COMDEX as they are going there. Hopefully, they boost things a bit from the regular 'year and launch'. After all, we are talking just about new revision of old chip in this case.

so Type def, you aren't alone.

 

[efty]

How can you guys compare something you have never seen?? Who has seen all the boards in action, side by side for comparison? This is ridiculous and IMO pointless. Even screenshots are not indicative of the quality of AA since they are static.

 

[Lincoln Bauman]

One thing I noticed about the 9700's AA is that it does not do a great job on high contrast edges.
This screenshot is with 6xAA, the bottom edge is the most noteworthy.
Looking at where the gradient starts, it goes from almost black to light gray.
The following gradients are very close in color, reducing the effectiveness of the antialiasing.
If it were represented with numbers, it looks more like "10 5 4 3 2" rather than "10 8 6 4 2".
In my opinion the "smoothness" of the gradient could be improved.

Lincoln