Anti-Aliasing/Anisotropic Filtering not working properly
Posted 07 July 2012 - 06:01 AM
Posted 07 July 2012 - 06:16 AM
Posted 07 July 2012 - 06:23 AM
Posted 07 July 2012 - 06:32 AM
Posted 07 July 2012 - 06:54 AM
Posted 07 July 2012 - 07:43 AM
Posted 07 July 2012 - 05:51 PM
I agree, SSAA + SGSSAA is overkill. SGSSAA is already applying to the entire scene post process, so MSAA is sufficient for AA for some initial cleanup of polygon edges.
Wow, HDR gets disabled in SLi? WTF?!? do you have a link for that?
and why are you running 4x4 supersampling? that's gonna VERY quickly eat up your VRAM since it has to render the entire screen 4 times your resolution. like for 1080p it has to render at 4320p, which I don't even think SLi can render that far without hitting the pixel clock limit so it might be disabling it altogether when you start the game. and especially if you're using TrSSAA or SGSSAA you shouldn't need anything quite that insane, regular MSAA should catch the stuff that TrSSAA doesn't. (also, SGSSAA adds a full screen blur to everything).
However, SGSSAA does not cause blurring of a scene by itself. It always comes down to the game and whether its post processing effects aren't compatible. Skyrim today is very compatible, and I get absolutely no blurring whatsoever.
Correction. TrSSAA needs to be used with SSAA, and TrMSAA needs to be used with MSAA. That is where they each rely on their repsective data, and using anything than those pairings will not provide the quality you are expecting. But you are right, they should be matched in power.
I also read somewhere that it's best to match MSAA with TrSSAA because TrSSAA gets it's data from the MSAA. so 4x with 4x or 8x with 8x, etc etc. but it's still possible to use more MSAA, just that the TrSSAA won't be quite as accurate at finding the nearest neighboring pixels and such.
Posted 07 July 2012 - 07:33 PM
Please keep in mind that whenever you use SGSSAA you should always match the number of MSAA samples used to the number of SGSSAA samples used. 8xMSAA + 4xSGSSAA will usually have lower quality than 4xMSAA + 4xSGSSAA. This is because SGSSAA gets its sample coordinates from the MSAA samples. So if you have more MSAA samples than SGSSAA samples the SGSSAA samples will not be placed in the ideal locations. Also you cannot have more SGSSAA samples than MSAA samples for the same reason (SGSSAA uses the MSAA sample coordinates). Even though it uses MSAA sample coordinates using additional coverage samples will reduce the effectiveness of SGSSAA by changing the grid pattern so that the MSAA samples are placed closer to the center of the pixel (which is a less ideal location). This is why I advised 4xMSAA + 4xSGSSAA and not 16xCSAA (4xMSAA + 12 extra coverage samples) + 4xSGSSAA.
there's also settings for LOD Bias depending on TrSSAA/SGSSAA/OGSSAA.
so you may be right about TrSSAA getting it's samples from SSAA, but it looks like SGSSAA uses MSAA for it's samples
Posted 07 July 2012 - 07:50 PM
Posted 07 July 2012 - 08:55 PM
Posted 07 July 2012 - 09:07 PM
BTW I just tested SGSSAA vs TrSSAA and you're right the SGSSAA doesn't blur in skyrim. and SGSSAA seemed to do a better job, it even reduced the shimmering on bump mapped textures (like the little ring-like things around the blessing stone thingies)
The thing to keep in mind is that SGSSAA is a form of SSAA, and is applied to the full screen post process, but also doesn't incur the performance hit that pure SSAA does. TrSSAA only works on alphatextures.
Posted 08 July 2012 - 10:05 AM
So you don't use any other settings in Nvidia Inspector? Just standard NVidia provided bits?
LOD adjustment is also being mis-represented on a lot of websites. You do not always have to adjust it, particularly when using SGSSAA. A lot of it comes down to the compatibility bits being used. Lowering also tends to introduce more texture shimmering. I'm using the Skryim bits provided by the latest drivers, LOD set to 0, and I have an outstanding picture with no texture shimmering.
Posted 08 July 2012 - 10:48 AM
0 user(s) are reading this topic
0 members, 0 guests, 0 anonymous users