Anti-Aliasing does not function

Should be here with recent drivers:
dsr-nvidia-control-panel-scaling-factors-640px.jpg


http://www.geforce.com/hardware/technology/dsr/technology
 

MACMAN86

Banned
Using Nvidia Experience? It's there in updated driver for down scaling from a much higher resolution. Thus smoother lines with no need for AA.
 
As I recall them stating they can not really tweak the AA. All they can do is make it for you can enable a switch that flicks a switch for your GFX card. In the end it really comes down to your GFX Card and drivers. They can not do much to change it. Sure they can implement some DX10/11/etc features that can give an illusion to fixing the AA but it would not really do it.

It would be nice to have the multiplier setting in-game though, not sure it will help much. The biggest problem comes from the design of the space stations. As you may have noticed they are really the only problems with jaggies once AA is on. Also the best AA seems to be SMAA.
 
Yes, the AA is a problem for me too. To the point where I decided to work extra hours to get a GTX 980 after I saw how DSR looks like on my GTX 750 Ti. I'm guessing I won't get constant 60fps with full DSR, but I'll try to push it as far as I can.
I would love better AA in the game, though.
 
YEs it really does just show itself in and around space stations but it is pronounced and needs to be fixed somehow. SweetFX and DSR are just band aids.
 
I think I have identified a cause of Aliasing amplification

Some of the surfaces on stations seem to have clear light to dark transitions when looking at the surface face on - on the second triangle in each quad. If you look at some of the aliasing, its not even aliasing, certain surfaces go from light to dark repeatedly over a 100 pixel area typically in a station, this to me is caused by the same thing, some models have an inverted vertex order/flipped normal, and this may be massively amplifying the impact of the aliasing.

If I find a station that looks the same again I will post a screenshot, it may not happen on all settings, but it seems to correlate with amplified aliasing, lighting flickering from off-to-on, and shadowing/lighting glitches on occasions with certain settings.
 
As I recall them stating they can not really tweak the AA. All they can do is make it for you can enable a switch that flicks a switch for your GFX card. In the end it really comes down to your GFX Card and drivers. They can not do much to change it. Sure they can implement some DX10/11/etc features that can give an illusion to fixing the AA but it would not really do it.
This is inaccurate: all our postprocess AA is handled by our own code, so we're not driver/manufacturer dependent. Multisampling is something we want, but under deferred shading (which we use) it's not a simple task. There are no DX10/11 features that would pretend to fix the AA, though handling the MSAA efficiently is most likely a DX11-only feature.

I think I have identified a cause of Aliasing amplification

Some of the surfaces on stations seem to have clear light to dark transitions when looking at the surface face on - on the second triangle in each quad. If you look at some of the aliasing, its not even aliasing, certain surfaces go from light to dark repeatedly over a 100 pixel area typically in a station, this to me is caused by the same thing, some models have an inverted vertex order/flipped normal, and this may be massively amplifying the impact of the aliasing.

If I find a station that looks the same again I will post a screenshot, it may not happen on all settings, but it seems to correlate with amplified aliasing, lighting flickering from off-to-on, and shadowing/lighting glitches on occasions with certain settings.
This is relevant to my interests. I can't quite visualise what you're describing but seeing screenshots of it would no doubt give us something to go on.
 
Bottom line Ben?

AA is shoddy.

Any plans to re-write or adjust the code to better make use of AA tech? or.....................is where we are now purely as a result of your writing for OR?

Thanks for a great product anyway!
 
Last edited:
This is inaccurate: all our postprocess AA is handled by our own code, so we're not driver/manufacturer dependent. Multisampling is something we want, but under deferred shading (which we use) it's not a simple task. There are no DX10/11 features that would pretend to fix the AA, though handling the MSAA efficiently is most likely a DX11-only feature.


This is relevant to my interests. I can't quite visualise what you're describing but seeing screenshots of it would no doubt give us something to go on.

@Ben Great to see you have found this topic. Lots of examples have already been posted, and I bet you do have an Oculus in the office with which this crawling behaviour is extremely obvious in station when you tilt your head. Do you still need more screenshots, or were you referring to another problem? Could you comment on the crawling issue as well in that case?
 
Elite is using deferred rendering (as do most games these days) to get cheap and good looking lighting/shading, but that also means that all hardware-antialiasing that's currently provided by graphics cards is unusable with the game, hence the need to do shader-based antialiasing. (If they weren't using deferred rendering the lights and shadows in the game would suck, so there'd also be complaining; probably even more than about the AA...)

Shader-based antialiasing is analysing the final image to find high contrast edges and smooth them, but since the game has many thin edges and surfaces that are one pixel or less thick when rendered at a shallow angle or from afar (like the mailbox grates) the rendered image itself contains discontinuities between parts of a single triangle/quad, and there's no shader-AA-algorithm that would be able to re-connect parts of a line that is broken up for several pixels.

Hardware based anti-aliasing knows where the edges of the geometry are (since it can be applied while the hardware is actually rendering the triangles) and can try to compensate for something like this, but shader-based AA doesn't have that luxury; the best mitigation you can currently have is upping the render resolution so the likelihood of features becoming too thin is lower.

So if there's less aliasing in other games it's either because they use hardware-anti-aliasing (though that's getting rarer) or because their geometry consists of less thin triangles and/or softer lighting with less stark brightness changes (like you'll usually get where there is an atmosphere, i.e. not in outer space).
 
^^^ Basically everything Leak just said.

The only thing Leak doesn't mention there is that (a) hardware multisampling data can be used in deferred shading if you write the extra code, sadly if you want efficient GPU scheduling you probably want tile-based deferred rendering (like wot Frostbite does) and (b) some games temporally antialias their environments because their environments aren't constantly moving. I believe Black Flag doesn't apply temporal AA to its characters for exactly that reason but I may be wrong. (c) Whatever Maiakaat described may be an art bug rather than a tech limitation.

Oculus specific: There's a distortion applied after scene rendering to make things the right shape for Oculus, naturally this is going to magnify some areas of the screen and minify others, it doesn't surprise me that any aliasing we have would be more visible in the magnified areas. Downsampling is probably going to be the best solution to this, in hand-wavey terms one could imagine a multi-res system that only oversamples in areas that it knows will be enlarged, but I'm talking off the top of my head here.
 
^^^ Basically everything Leak just said.

The only thing Leak doesn't mention there is that (a) hardware multisampling data can be used in deferred shading if you write the extra code, sadly if you want efficient GPU scheduling you probably want tile-based deferred rendering (like wot Frostbite does) and (b) some games temporally antialias their environments because their environments aren't constantly moving. I believe Black Flag doesn't apply temporal AA to its characters for exactly that reason but I may be wrong. (c) Whatever Maiakaat described may be an art bug rather than a tech limitation.

Oculus specific: There's a distortion applied after scene rendering to make things the right shape for Oculus, naturally this is going to magnify some areas of the screen and minify others, it doesn't surprise me that any aliasing we have would be more visible in the magnified areas. Downsampling is probably going to be the best solution to this, in hand-wavey terms one could imagine a multi-res system that only oversamples in areas that it knows will be enlarged, but I'm talking off the top of my head here.

Hey, traduce :D

better FXAA or SMAA to GTX 560 :) ?
 
I have tried all the AA options and the difference is that small it's near undetectable. (unattractive jaggies/shimmering on sharp lines). In short, the AA in ED is no good.

GTX 770 with the latest drivers. It was the same with all previous drivers.
 
^^^ Basically everything Leak just said.

The only thing Leak doesn't mention there is that (a) hardware multisampling data can be used in deferred shading if you write the extra code, sadly if you want efficient GPU scheduling you probably want tile-based deferred rendering (like wot Frostbite does) and (b) some games temporally antialias their environments because their environments aren't constantly moving. I believe Black Flag doesn't apply temporal AA to its characters for exactly that reason but I may be wrong. (c) Whatever Maiakaat described may be an art bug rather than a tech limitation.

Oculus specific: There's a distortion applied after scene rendering to make things the right shape for Oculus, naturally this is going to magnify some areas of the screen and minify others, it doesn't surprise me that any aliasing we have would be more visible in the magnified areas. Downsampling is probably going to be the best solution to this, in hand-wavey terms one could imagine a multi-res system that only oversamples in areas that it knows will be enlarged, but I'm talking off the top of my head here.

It would only benefit people with Maxwell cards but would MFAA help? It's still hardware based so there is that performance cost but it isn't as cost intensive as full MSAA
 
Back
Top Bottom