Shimmering

[Stringy]

Well we threw around the big words and acronyms pretty well in this thread...
I knew what SuperSampling was, I knew MultiSampling was better than SuperSampling... but not how Multisampling gets the results... don't really care though either...
Don't know who does what, and still don't [as far as nV and ATi]...

I did find out that the P does help shimmering, but the evidence is "you have to see it"
which is true, once we see it, we know it solves this, but some just like the reassurance of it being explained Before the purchase an expensive peice of equipment... I know I'm difficult...

So really I see it as we must define what the problem the FAA takes care of.. [1]the "Shimmering" which is the Alaising on the horizon and/or [2] the Texture shifting as I like to call as 2 textures go front to back fighting for the same position on a "Tiled" scene, the most common terrain format used in Flight Sims...

So, FAA as done with the P, will take care of the Alaising in the horizon, because it aliases the edges by its purpose...
But [and here is a Q] Does this also mean it will stop texture swimming? or what-ever you wanna call it today...



Craig

[Ryu Connor]

If so can you not ask the person with the Parhelia to try out a flight sim using 16xFAA?

Scott doesn't own a flight sim.

http://www.simhq.com/

- SimHQ Matrox Announcement: Matrox has lifted their NDA on Parhelia-512. We have a sample in-house and are working quickly to finish Part 1 of our review. Stay tuned for a probable Part 1 tonight. In the mean time, be sure to revisit our preview of Parhelia-512 right here.

This is the site to check out. Shouldn't be too much longer I suspect.

[Greebe]

Why would Ryu need to McElvis? Rags has already stated that 16x FAA does stop the shimmering in M$ Flight Sim.

[McElvis]

Well, then he can see for himself

[Greebe]

Gotcha

[Ryu Connor]

I'll be able to tell for myself shortly after the seventh.

[xortam]

Personally, I appreciate the time you're taking to explain your thoughts and include citations Ryu Connor. I haven't dealt with most of these issues with my simple 3D experience and my Max but I do appreciate learning more about the subject. I'll archive this thread and scour through it and its citations more thoroughly later. I think that ultimately most people would only care about comparing the final results and not comparing the various algorithms that may or may not be currently correctly implemented, i.e. if it looks good and performs well, I don't care how they did it.

[[Ch]amsalot]

Some of at leat <my> confusion came over the actual definition of multisampling vs. supersampling.

It turns out that many tech sites as well as 3dfx at one time called Voodoo5's FSAA "multisampling". Now it appears that it's actually supersampling by taking samples from a rotated grid.

Also, what I previously thought was supersampling by use of a larger version of the scene was actually supersampling by means of taking samples from an "oversampling" (enlarging) the original scene. Confused yet??

I have yet to understand the definition of multisampling. It was a horrible day for me at the Jeep dealership. I'll try to figure this all out tomorrow.


-[Ch]amsalot

[xortam]

multisampling vs. supersampling

I think anybody that read this thread through probably understood that [Ch]amsalot.

I recall when I was developing new architectures that one of the big early tasks was defining terminology so that we could all clearly communicate with each other. Poorly defined and/or misused terminology can confuse a lot of people and cause a great deal of wasted time.

[jkakaley]

As I read this forum I see a lot of misinformed statements so I'll try and give you some insight into what I know about antialiasing.

First off, the quotes made by hardware review sites like techreport, etc. don't mean anything. While they do come in contact with the hardware almost every day, I'd put my faith in the white papers published by the companies that actually manufactured the cards like 3dfx, nVidia, and Matrox.

Now for the 3dfx method. The Voodoo5 renders the scene at a higher resolution to an off-screne display buffer then "jitters" the pixels (hence Rotated Grid Antialiasing) to blend them together and scales the image back down to what the display resolution actually is. As everyone with a Voodoo5 card will tell you, the Image quality is great and no texture shimmering occurs, but because of it's bluring method, you lose a slight amount of sharpness in the textures, but you get no texture shimmering as the tradeoff.

Now the method for the GeForce2 and below. It is basically the same as the Voodoo5 except it dosen't have any jitter effects to blend the pixels together, hence, it looks like crap compared to 3dfx and you basically get a tiny bit more image quality at a rediculous framerate cost.

In the GeForce3 the used an effect called multisampeling which still renders the ENTIRE image at a higher resolution into an offscrene buffer, expect it "shares" sub-samples between several different pixels which does two things. One, the shareing of pixels will blend the image so the image quality is a lot better, and two, the shareing of pixels means that less pixels will have to be rendered, thus increasing performance. This method still dosen't offer the quality that 3dfx does because 3dfx uses more pixels in its blending algorithm, but it is a large improvement over the GeForce 2.

In the GeForce4, nVIdia took the multisampling algorithm and made some adjustments as to "which pixels where shared with which", and as a result, blured the textures less, but blured the edged together the same. The ENTIRE scene was still multisampled, but the algorithm had the biggest effect on the edges.

Matrox uses a totally different approach, and detects which pixels lie on an edge and supersamples them at 16 times the resolution. I don't know if the have a special blending algorithm like "jittering" or multisampeling, but I do know regardless, at 16 times the resolution, a LOT of the jaggies will go away. This method dosen't touch the textures, so, any shimmering artifacts in the textures will not be removed.

One more word about texture shimmering. I have noticed that it occurs in old games mostly, and I think that it is because older games had low-resolution textures and that meant the textures had to be "stretched" over the polygons and I think that causes shimmering. I have heard that anisotropic filtering reduces the shimmering effect, but in my experience with a Radeon 8500 (my current card), it enhances the texture quality greatly, but sometimes it causes more shimmering than before, so I guess it depends on the aniso method.

[ALBPM]

Sniff- sniff- I do believe I smell something burning around here.....

Any Forest Service Employees around here....or what?????

[jonhart]

Oooo, let me try...

The different antialiasing techniques are quite confusing, but we seem to be nearly there.

Im pretty sure that the voodoo doesnt store a large offscreen buffer to be sampled down ( like the geforce and radeon do ), but uses the parallel architecture of its VSA chips to process each pixel in parallel, and combine the results.
The rotated grid is determined by the pixel center settings of each chip ( it could be ordered grid if the chips were programmed that way ). It is quite elegant really.

Multisampling is quite a neat idea as well. The scene is rendered at, say, 4x resolution. But each time a pixel is written, it use the same texel as its neighbours. This doesnt seem to gain you anything, but each pixel gets its own z buffer entry. If another polygon happens to overwrite the area, then some of those 4 pixels will have different values. When the pixels are all combined to reduce the resolution the ones that got overwritten will get a n averaged final result, while those that weren't will stay the same colour. This has the effect of only anti aliasing edges. But it still requires a lot of bandwidth to write all those pixels out.

FSAA is really the elegant solution, as it only uses bandwidth when it needs it. Its not original ( it is a well known technique in raytracing ), and so I think we will see it become commonly used on all products. The onlu drawback is that because the scene isnt held in memory at high resolution multi pass techniques ( like stencil shadows ) dont get anti aliased against previous passes ( probably applies to the voodoo parallel technique as well ).

In the end there are techniques used by every manufacturer that we would like to see combined into one mega card. Geforce memory technology, matrox anti aliasing, radeon aniso filtering, kyro occlusion culling, 3dlabs fine granularity.

Personally, Im thinking about a parhelia because of dual DVI, and FSAA.

[Helevitia]

Hell, I'll give it a shot....

You have this edge, and it is bumpy, like stairs. We don't like bumpy edged stairs on our flight sims, do we? So we take a big hammer and....oh nevermind, it'll be to technical for you guys

I think the heat is effecting me. I'm starting to liek the effect though.

Dave

[The PIT]

Originally posted by jkakaley
[B]
First off, the quotes made by hardware review sites like techreport, etc. don't mean anything. While they do come in contact with the hardware almost every day, I'd put my faith in the white papers published by the companies that actually manufactured the cards like 3dfx, nVidia, and Matrox.
B]

Yep I agree with that statement.

[Ryu Connor]

I have heard that anisotropic filtering reduces the shimmering effect, but in my experience with a Radeon 8500 (my current card), it enhances the texture quality greatly, but sometimes it causes more shimmering than before, so I guess it depends on the aniso method.

ATI doesn't apply AF in a uniform fashion.

JonHart:

I think you mean FAA. Saying FSAA will just create further confusion.