Help wanted for Supersampling Antialiasing testing

[Ross Scott]

I've been a junkie about antialiasing ever since the feature first emerged in videocards. I love how it cleans up the image and try to enable it in every 3D thing I do. A lot of more modern games however either don't support antialiasing at ALL or else have really spotty coverage so you have what's called "shader aliasing", where some edges look perfectly clean, but others with special light shader effects look pixellated and ignore AA settings. In these cases, sometimes supersampling antialiasing (SSAA) is the only way to clean up an image, even if it takes a massive performance hit.

 

Both AMD and Nvidia offer different SSAA settings, but I've only seen a minimal amount written on the topic and poor documentation on it in general. I'd like to get more information on which cards are more compatible, look better, and run faster for SSAA. Some cards make be more compatible, but cause blurring, others may take a massive speed hit, who knows.

 

If you have a Geforce 400 / 500 series card, or a Radeon 5000 / 6000 series card, you can help out. You can do this in one of the following ways:

 

 

Method 1: Find an area in a game you have where "normal" multisampling or MSAA/SSAA combo AA doesn't work well and you can still see a CLEAR "stairstep" look to some edges. Most games based off the Unreal 3 engine are like this. Take a screenshot of that, then enable pure SSAA and take another screenshot, to show the difference. Please don't post stand-alone screenshots here if you don't have anything to compare them to.

 

 

Method 2: Make a benchmark of one of your games in the following modes:

 

A. No AA being used

B. "normal" 4x AA being used

C. 4x4 SSAA being used

 

Be sure to state what videocard you have along with how much memory it has (the rest of your specs isn't super important, since I'm interested in the difference between the modes, not overall performance). Be sure to keep other detail settings the same for each benchmark.

 

Also a warning that 4x4 SSAA can be EXTREMELY slow. You may change your mind about how fast your computer is once you try this out. If the game is suddenly massively slower, don't worry, that means it's working.

 

 

Method 3: List any games you have where antialiasing simply will NOT work for you. Or list games where SSAA works, but it causes the image to be very blurry (Half-Life 2 Episode 2 is an example of this).

 

 

The more information I get, the better. It will probably influence the next videocard I buy. Thanks in advance to anyone who contributes information.

 

- - -

HOW TO ENABLE SUPERSAMPLING

 

For Nvidia users:

 

Download the program "Nvidia Inspector" then change the following settings for the game:

 

 

 

 

If you see any settings related to "rotated grid" definitely mention them here. I currently have a Geforce 9800 GTX, so I may not have as many options as you.

 

 

For ATI/AMD users:

 

Go to the Catalyst Control center under AA, and find a slider towards the bottom:

 

 

Change it to "Super-Sample AA." I recommend doing it in 4x mode instead of 8x since then tends to hit a good balance between picture fidelity and too much blurring. Also to the best of my knowledge, ATI's SSAA only works in DirectX 9 mode or lower (not 10 or 11).

[Alyxx Thorne]

Here's Fallout 3 GOTY with three different modes at 1280x720. I noticed no notable slowdown using 4x SSAA, though my computer has a 3.2 GHz quad CPU and 4GB high speed RAM in it. I noticed a much smoother image using SSAA than normal AA actually.

 

First, some info about my card:

 

XFX Radeon HD 5770 with 1GB VRAM (XFX being a cheaper version but essentially it's an ATI card).

 

 

And here's the screenshots, taken outside the gate of Megaton which I consider the easiest place to spot AA issues for obvious reasons. Click the thumbnails to view full version.

 

No AA:

 

4x AA:

 

4x SSAA:

 

I might upload similar screenshots of Bulletstorm later on which uses the Unreal 3 engine.

[Ross Scott]

Hmm, well thanks for the screenshots, but I'm not sure it's working in Fallout 3 for you. SSAA gets rid of basically ALL jagged edges, but I'm not seeing much of a difference between that and regular MSAA from your shot. Look at the bar sticking out from ground next to the chair on the right. It's jagged in both AA shots. If you want to be certain it's working, you might try turning off in-game AA, but keep it enabled from the control panel and see if it still works. If it doesn't, it means that for whatever reason it's not working in Fallout 3 for you. Benchmarks are another quick test. If the framerate is about the same between both AA modes, it means it's not working. Even on the best hardware SSAA causes a big framerate drop. You may have better luck with bulletstorm however.

 

Here's an example of what to expect for SSAA mode (though hopefully not this blurry in most games):

 

http://www.hardocp.com/image.html?image=MTI1MzU4OTM1NVlDbXBla3ZKZm5fNl8xMF9sLnBuZw==

[Alyxx Thorne]

Hmm, I cannot seem to get SSAA working, even if I attempt to override the game settings...

[Ross Scott]

I'll consider that one game SSAA doesn't work on for Radeon cards for now then ( I can try and take a screenshot on my Geforce sometime later). Definitely give the Unreal 3 engine games a shot however.

 

EDIT: I forgot to mention that I think the Radeon's SSAA only works in DirectX 9 mode. I don't know if Fallout 3 is DirectX 10 or not, but if so, forcing it 9 might get it working.

[BTGBullseye]

I've got a Radeon 5470, runs with shitty FPS when SSAA is on (mobile card with only 80 Stream Processors), but I can probly get some SS of Mass Effect, Alien Swarm, HL2 (& episodes), GTA4, Battlefield 2142, Battlefield: Bad Company 2, Far Cry 2, Medal of Honor, HL:S (I usually run it with SSAA 4x, and Wide Tent AA Filtering), Dead Space 2, Rainbow 6: Vegas 2, Assassin's Creed (1, 2, and Brotherhood), Star Wars: Empire At War, all of the Command & Conquer games (except 4), Tropico 3, Civilization V, Dragon Age 2, and Warzone 2100.

 

Let me know what SS you want, what AA filter settings you want (Box, Narrow Tent, Wide Tent, Edge Detect), and what different AA levels you want... (2x, 4x, at MSAA/AMSAA/SSAA)

 

If you want the best cards for the best AA at the best speeds, GeForece 5xx series beats all, followed closely by the 4xx series, then the top of the line Radeons (6900 series), then the GTX 285, then the Radeon 6800 series... (I'm a hardware buff, I can find hardware specs and performances for anything)

 

For overall GFX card performance, check here: http://www.videocardbenchmark.net/

[Ross Scott]

Well that's a lot of games to cover, I'll look over all screenshots / benchmarks you're willing to send however. I'll be surprised if Dead Space 2 in particular works, the original one was notorious for not working with ANY AA modes. The ONLY method that works on that at this time is ATI's MLAA mode, which is a post-processing technique rather than "real" AA (still better than nothing though).

 

As for box, wide tent, etc. I THINK by enabling SSAA, that will override all those modes, so it won't matter. The trick is to find games or areas in games where "normal" AA modes like the ones you mentioned still don't smooth all the jagged edges due to shaders or whatever. For instance, HL2 isn't a great game for that, since regular AA modes clean up so much of the image as it is. Anything based on Unreal 3 is particularly good for finding shader aliasing however. I'd say run your settings at 4x for comparison (I never notice huge differences beyond 4x), pick a filtering mode you want, maybe put anisotropic filtering to 16x.

 

Finally, my guess is SSAA performance is equally punitive on both cards, but the big question I've never had answered is how big a difference the amount of video memory makes. Take the 2GB Geforce 460 for example. In 99% of applications that most people run, the extra 1GB of memory makes NO difference in performance. But for something like supersampling, it MIGHT, I don't know; I've never seen a comparison or know just how much video memory a card can really use for SSAA. The very nature of supersampling suggests that it could use a LOT of memory. Also for some games apparently the SSAA image quality on Nvidia cards can be blurrier, but I've barely seen ANY information on the topic to make an informed opinion on it.

[BTGBullseye]

Well that's a lot of games to cover, I'll look over all screenshots / benchmarks you're willing to send however. I'll be surprised if Dead Space 2 in particular works, the original one was notorious for not working with ANY AA modes. The ONLY method that works on that at this time is ATI's MLAA mode, which is a post-processing technique rather than "real" AA (still better than nothing though).

I'll have to see, I never use AA on the newer games because it seems to drop my framerate to somewhere between 5 and 15 FPS...

 

As for box, wide tent, etc. I THINK by enabling SSAA, that will override all those modes, so it won't matter. The trick is to find games or areas in games where "normal" AA modes like the ones you mentioned still don't smooth all the jagged edges due to shaders or whatever. For instance, HL2 isn't a great game for that, since regular AA modes clean up so much of the image as it is. Anything based on Unreal 3 is particularly good for finding shader aliasing however. I'd say run your settings at 4x for comparison (I never notice huge differences beyond 4x), pick a filtering mode you want, maybe put anisotropic filtering to 16x.

I'll use 16x Anisotropic, but the different filtering does still make a big difference with SSAA. I'll give you what I got from the Catalyst Control Center help file...

• Box or Standard—Uses only samples from within the current pixel. All samples are equally weighted. (Produces the sharpest look; fastest performance.)
  
• Narrow-tent—Uses samples from the current pixel and the surrounding area. Samples are weighted based on their distance from the pixel center. (Produces a softer look; slightly slower performance.)
  
• Wide-tent—Uses samples from the current pixel and the surrounding area. Samples chosen by the wide-tent filter include areas a little bit further away from the current pixel than samples used by the narrow filter, but not from areas further away than 1.25 times the current pixel radius. Samples are weighted based on their distance from the pixel center. (Produces a softer look; slightly slower performance.)
  
• Edge-detect—Uses samples from the current pixel and the surrounding area. Samples chosen by the edge detect filter are weighted to produce better anti-aliasing along edges without causing any softening or blurring of the scene. (Slightly slower performance.)

 

Finally, my guess is SSAA performance is equally punitive on both cards, but the big question I've never had answered is how big a difference the amount of video memory makes. Take the 2GB Geforce 460 for example. In 99% of applications that most people run, the extra 1GB of memory makes NO difference in performance. But for something like supersampling, it MIGHT, I don't know; I've never seen a comparison or know just how much video memory a card can really use for SSAA. The very nature of supersampling suggests that it could use a LOT of memory. Also for some games apparently the SSAA image quality on Nvidia cards can be blurrier, but I've barely seen ANY information on the topic to make an informed opinion on it.

I'm again going to quote from the help file...

 

MSAA renders 3D objects by rendering multiple samples at various programmable locations within each edge pixel and blending the sample values together using gamma-correction, smoothing the appearance of polygon edges. SSAA works similarly, but renders multiple samples for every pixel on the screen, unlike MSAA which only renders multiple samples at polygon edges. By rendering multiple samples for every pixel, super-sampling gives a higher quality image than multi-sampling, reducing shimmering in textures and shading as well as at polygon edges. Adaptive anti-aliasing uses a combination of both methods: SSAA to render some surfaces that have transparent elements and MSAA for all other surface rendering.

 

Super-sampling can be much slower than multi-sampling because it effectively renders the whole scene at a much higher resolution than the display. A 2560×1600 image with 4× super-sampling takes about the same amount of time to render a 5120×3200 image without super-sampling.

Adaptive MSAA is essentially SSAA on transparent areas, and MSAA for everything else...

 

Ok, so the short of that last bit is: Whatever the multiple is (2x, 4x, 8x, 16x, etc.) is multiplied to the resolution of the textures/scene and then down converted. (MSAA is only for 3d objects)

 

 

The MSAA/SSAA setting says what gets the rez boost, the filtering tells how many pixels to sample, and from where.

[Ross Scott]

I'll use 16x Anisotropic, but the different filtering does still make a big difference with SSAA. I'll give you what I got from the Catalyst Control Center help file...

None of the filtering methods mentioned are SSAA however. While it's structured differently from the other options in the control panel, I believe selecting SSAA overrides other AA modes, that's how it works on Nvidia cards at least. ATI's could be structured differently so that it does one full AA mode on top of another, but that wouldn't make a lot of sense since I think enabling any AA at all picks one of the modes mentioned by default. I could be wrong on this, but I'd need some solid proof established first (if you can get notable benchmark differences between the different modes WITHOUT ssaa, that would be the first step, we can test that as well.) Right now I think saying it makes a big difference with SSAA is like saying wide tent makes a big different with narrow tent; it doesn't, it's a completely separate mode.

 

Adaptive MSAA is essentially SSAA on transparent areas, and MSAA for everything else...

I'm quite familiar with Adaptive AA, I used it in some of the earlier CP episodes even. It was a badly needed mode when it came out back in 2005. You'd think it would be the best of all worlds, but unfortunately it doesn't do anything for shader aliasing or deferred rendering, which are increasingly common in some games now.

 

Ok, so the short of that last bit is: Whatever the multiple is (2x, 4x, 8x, 16x, etc.) is multiplied to the resolution of the textures/scene and then down converted. (MSAA is only for 3d objects)

Dude, I understand how it works, but game-to-game image quality, benchmark differences versus video memory and chipset manufacturers, and flat out game COMPATIBILITY are all big unknowns to me. The help file isn't even completely accurate. If everything worked perfectly, SSAA would always give better image quality, but as it stands, bugs exist. Try running HL2 Episode 2 with SSAA v. MSAA and see how much blurrier it is. In those cases, the quality is inferior, it's a real tossup between Nvidia, ATI (AMD) and the games being run.

[BTGBullseye]

Ok, so I've got some from these games:

 

• Fallout: New Vegas
  
• Half-Life 2 (this one was the hardest, read the .txt file for the why)
  
• Half-Life 2 w/ Cinematic Mod
  
• Half-Life: Episode 2 w/ Cinematic Mod
  
• Half-Life: Source w/ HD pack
  
• Rainbow Six: Vegas 2
  
• Star Wars: Empire At War

 

They're split into their own sub-folders for easier browsing, and comparison. All SS are in .png format. The total size of the .rar is 69.7 MB, so I'm putting it in a torrent. I'll be seeding for quite a while...

 

Let me know if you have questions, if you want more SS for any of these, or if you want SS of another game... (I can likely torrent any game you want to see SS of that I don't already have)

[Ross Scott]

Hey,

 

Well I'm definitely interested in seeing the comparisons, however a torrent isn't the best option when you're the only host. Would you mind uploading them to someplace like www.megaupload.com, 4shared, or some other free hosting site instead?

[BTGBullseye]

lol, ok... I'll keep seeding the torrent though. lol

 

Here: http://www.megaupload.com/?d=TXZ4RKHA

 

BTW, I have a 3Mbps up rate...

[Ross Scott]

Thanks for the pictures, it's interesting how SSAA seems to produce the best image quality in some games, but causes more blurring than fixes on others. I wouldn't mind seeing some shots of Mass Effect (any area that's loaded with shader aliasing) especially, or additional shots of Rainbow Six Vegas in areas where there is shader aliasing or deferred rendering (your existing test was good, the gun rack in the background obviously has it). I'd say you don't have to bother with every mode however. I think 2 shots would be fine. Maybe one with 4x box AA / 16x AF and one with 4x SSAA, still box.

[BTGBullseye]

Ok, I'll do a couple shots ASAP...