GeForce GTX 460 SLI vs. Radeon HD 5970: Two Against One

Page 5 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Wreckage

Banned
Jul 1, 2005
5,529
0
0
I don't think that's invalid. The point is: there are plenty of IQ enhancements that we don't have *yet* in real-time, which are more important than resolution.



Apparently we don't share the same opinion on what makes something 'realistic'.
I would much prefer the added geometry detail, animation, physics and AI over Far Cry, at the cost of a slightly lower resolution and a bit of aliasing.

Far Cry at 2650x1600 with SSAA would just be a very clinically rendered world that isn't very realistic. Reminds me too much of the raytracing crowd... "Look at this beautiful collection of perfectly round spheres on a checkerboard that I've rendered!". Yea okay... but we don't have all that much in the form of spherical objects and checkerboards in the real world. So it may be cleanly rendered, but it's not a realistic 'world'.

I just picked up a 23" monitor and considering I sit at a desk, a much bigger monitor is not needed. So I completely agree with what you said. I think a lot of people are fine at 1280x1024 or 1650x1050. They want more eye candy to use on their existing monitors.

For example look at the recent steam survey, here are the top resolutions.
http://store.steampowered.com/hwsurvey/

Code:
1024 x 768
10.71%
0.00%

1152 x 864
1.87%
-0.03%

1280 x 720
0.75%
+0.05%

1280 x 768
0.78%
+0.02%

1280 x 800
7.36%
-0.26%

1280 x 960
1.33%
-0.08%

1280 x 1024
16.75%
-0.34%

1360 x 768
1.27%
+0.06%

1366 x 768
6.06%
+0.48%

1440 x 900
10.91%
-0.30%

1600 x 900
3.06%
+0.20%

1600 x 1200
0.85%
-0.03%

1680 x 1050
16.71%
-0.08%

1920 x 1080
11.84%
+0.75%

1920 x 1200
5.68%
-0.29%
 

MrK6

Diamond Member
Aug 9, 2004
4,458
4
81
Unfortunately, I think a lot of people who argue "oh, high resolution isn't the end all" most likely haven't experienced a 30" monitor @ 2560x1600 in all it's glory. And I mean I actually owned/borrowed one, used it for a month, and really experienced it. I have never met a single person who went high resolution who wanted to go back, it brings realism to a whole new level. There is nothing like it out there, no matter how many multi-monitor/3D/other marketing gimmicks companies try to throw out as competition.
 

Grooveriding

Diamond Member
Dec 25, 2008
9,110
1,260
126
I don't think that's invalid. The point is: there are plenty of IQ enhancements that we don't have *yet* in real-time, which are more important than resolution.



Apparently we don't share the same opinion on what makes something 'realistic'.
I would much prefer the added geometry detail, animation, physics and AI over Far Cry, at the cost of a slightly lower resolution and a bit of aliasing.

Far Cry at 2650x1600 with SSAA would just be a very clinically rendered world that isn't very realistic. Reminds me too much of the raytracing crowd... "Look at this beautiful collection of perfectly round spheres on a checkerboard that I've rendered!". Yea okay... but we don't have all that much in the form of spherical objects and checkerboards in the real world. So it may be cleanly rendered, but it's not a realistic 'world'.

Higher resolution is always the future going forward. I can remember not too long ago when 1920x1200 was considered huge, now a lot of people have it.

1680x1050 was then added to allow people to move up at a more affordable price point (TN monitors) and it is now the de facto gaming resolution for a lot of people.

It will continue on this way. Bigger resolutions will always become more standard.

If one could imagine a current 30" monitor with a resolution of say 5120x3200, that would be image quality nirvana. Densely packed pixels with a very small dpi, would make for an amazing picture quality.

Higher resolution = much better IQ. ATI and nvidia also are seeing higher resolutions as the future for them going forward, as currently there is too much gpu power for where most people are gaming at resolution wise. Hence why they are pushing multi-monitor gaming, they're trying to funnel gamers to situations that will make more demands gpu wise, and keep them upgrading.

That's also why we get garbage like physx that gives massive framerate hits for pretty much nothing gameplay wise.

The multi-monitor push to get gamers up to higher resolutions makes sense, as the price of three TN panels is $600 and takes you right up to 5760x1200 vs paying $1000+ for a 30" screen that only gets you to 2560x1600.

The problem with the multi-monitor thing is that it does not give you the IQ increase you get from a quality super large single display, if anything, it degrades IQ even more by adding in bezels and awkward setup situations.

Increased resolutions will continue to be a driving force in PQ/IQ. There are already plans to eventually bump up the HDTV standard from 1920x1080 to higher resolutions in the future. That may help to catalyze and alleviate costs in pc monitors also making a resolution leap vs price.

I wish I had the time and information to research more about LCD technology and costs and what is possible with it. I'd like to understand where the prohibitive aspects come in when it comes to increasing resolutions at a given panel size, and cost of production for various panel types and sizes. My assumption on 30" monitor prices, is that there just is not enough demand currently to drive the price down.
 
Last edited:

BFG10K

Lifer
Aug 14, 2000
22,709
2,995
126
I don't think that's invalid. The point is: there are plenty of IQ enhancements that we don't have *yet* in real-time, which are more important than resolution.
Right, that’s the point, and that’s why it’s invalid. Offline enhancements include massive levels of filtering and AA, which is why pre-rendered CGI movies don’t have a single pixel out of place. PhysX/Tessellation/SSAO won’t achieve that.

MSAA is almost useless for modern games with shader aliasing; we need SSAA derivatives for that, and that has a much higher performance hit.

Apparently we don't share the same opinion on what makes something 'realistic'.
I don’t consider an image rendered onto a swarm of bees (e.g. Crysis without super-sampling) to be realistic, and nobody would think those movies would look realistic if they shimmered like crazy too.
 

Scali

Banned
Dec 3, 2004
2,495
0
0
Right, that’s the point, and that’s why it’s invalid. Offline enhancements include massive levels of filtering and AA, which is why pre-rendered CGI movies don’t have a single pixel out of place. PhysX/Tessellation/SSAO won’t achieve that.

MSAA is almost useless for modern games with shader aliasing; we need SSAA derivatives for that, and that has a much higher performance hit.

I don't think you quite understand the implications of tessellation.
Do you know what causes shader aliasing? It's noise introduced by per-pixel bump/offset maps (poor-man's substitutes for geometry). MSAA cannot handle this because it only works on polygon edges.
Replace bump/offset mapping with tessellation, and what do you get? Exactly, you get polygons, not pixels. And that means that MSAA will correctly antialias, and SSAA will not offer any benefits.
 
Last edited:

BFG10K

Lifer
Aug 14, 2000
22,709
2,995
126
Do you know what causes shader aliasing?
Yes, a shader being undersampled. Or in signal theory terms, it’s high frequencies masquerading as low frequencies.

It's noise introduced by per-pixel bump/offset maps (poor-man's substitutes for geometry).
This is but one of many ways to generate shader aliasing.

Exactly, you get polygons, not pixels. And that means that MSAA will correctly antialias, and SSAA will not offer any benefits.
Those polygons still need shaders applied to their internals (e.g. lighting). Likewise, tessellation won’t affect textured surfaces. So unless you’re planning on having solid colors for the whole scene, you still need something better than MSAA.
 

Scali

Banned
Dec 3, 2004
2,495
0
0
Yes, a shader being undersampled. Or in signal theory terms, it’s high frequencies masquerading as low frequencies.

Yes, but since gradients and texture lookups are all filtered, these generally don't cause shader aliasing.
It's really mainly just normal mapping and related effects that simulate geometry, which introduce sampling issues. Filtering is already applied for everything that just follows the surface contours.

tessellation won’t affect textured surfaces. So unless you’re planning on having solid colors for the whole scene, you still need something better than MSAA.

Why do you think texture filtering exists?
In fact, SSAA is not exactly a good way to battle texture/colour aliasing issues. You need a LOT of supersampling to handle things that a normal anisotropic texture filter can handle in a single shader lookup.
 

SirPauly

Diamond Member
Apr 28, 2009
5,187
1
0
Unfortunately, I think a lot of people who argue "oh, high resolution isn't the end all" most likely haven't experienced a 30" monitor @ 2560x1600 in all it's glory. And I mean I actually owned/borrowed one, used it for a month, and really experienced it. I have never met a single person who went high resolution who wanted to go back, it brings realism to a whole new level. There is nothing like it out there, no matter how many multi-monitor/3D/other marketing gimmicks companies try to throw out as competition.

Here is a nice trick some should do then. Try a 24 inch 1900 x 1200 with no AA and compare it to 1680 x 1050 with 22 inches with a super-clean screen -- let's say, using a hybrid mode to clean texture and polygon aliasing and move around in the title.

Here is another ---- Try a 24 inch monitor with 2d gaming and use a 22 inch monitor in 3d stereo in a quality title that shines in 3d stereo. Here is an idea: Maybe the 3d stereo monitor considering it is offering the illusion of dimensions of depth and pop-out -- literally looks much larger based on the added dimensions.

Resolution is not the end-all-be-all but obviously it is a wonderful foundation to build upon to add things to.

Multi-monitor is a gimmick? But yet one could add three 30 inch monitors? Must be impressive to see the immersion factor of three monitors instead of just one.
 

SirPauly

Diamond Member
Apr 28, 2009
5,187
1
0
Why do you think texture filtering exists?
In fact, SSAA is not exactly a good way to battle texture/colour aliasing issues. You need a LOT of supersampling to handle things that a normal anisotropic texture filter can handle in a single shader lookup.

No, it's not, but considering the performance and the ability to go multi-GPU, it is like a sledge hammer instead of a graceful efficient feature. But, what it may do is curb aliasing issues that the developer didn't take great care or, tools simply don't exist yet. A stop gap till aliasing has been conquered, destroyed in an efficient manner to me.
 

Scali

Banned
Dec 3, 2004
2,495
0
0
No, it's not, but considering the performance and the ability to go multi-GPU, it is like a sledge hammer instead of a graceful efficient feature. But, what it may do is curb aliasing issues that the developer didn't take great care or, tools simply don't exist yet. A stop gap till aliasing has been conquered, destroyed in an efficient manner to me.

People using SSAA to try and cure texture aliasing always gives me a bad taste in my mouth.
Usually with (realtime) raytracing, as it's easier to implement an anisotropic filter in a rasterizer than in a raytracer, because of the access to gradients and neighbouring pixels.
And don't forget, this also includes mipmapping (just calculating a miplevel is not exactly trivial in a raytracer).
If you want to render everything with high-res textures, and no mipmapping... well, I think you get my idea about the EXTREME levels of supersampling you'll require to get every texel filtered nicely on screen.

I don't think texture filtering will ever be replaced by screen-space supersampling, because they just solve different problems.
 

SirPauly

Diamond Member
Apr 28, 2009
5,187
1
0
People using SSAA to try and cure texture aliasing always gives me a bad taste in my mouth.
Usually with (realtime) raytracing, as it's easier to implement an anisotropic filter in a rasterizer than in a raytracer, because of the access to gradients and neighbouring pixels.
And don't forget, this also includes mipmapping (just calculating a miplevel is not exactly trivial in a raytracer).
If you want to render everything with high-res textures, and no mipmapping... well, I think you get my idea about the EXTREME levels of supersampling you'll require to get every texel filtered nicely on screen.

I don't think texture filtering will ever be replaced by screen-space supersampling, because they just solve different problems.

I hear ya, but with a gamer context; what can I do? I have a title that exhibits texture aliasing -- have the extra performance -- and appreciate a super-sampled coupled with filtering option at times 'till developers take more care with their titles or technology evolves.
 

Scali

Banned
Dec 3, 2004
2,495
0
0
I hear ya, but with a gamer context; what can I do? I have a title that exhibits texture aliasing -- have the extra performance -- and appreciate a super-sampled coupled with filtering option at times 'till developers take more care with their titles or technology evolves.

Well yes, I see how you can reduce the aliasing in current games by bruteforcing SSAA on them... but that's only because these games are inherently flawed.
We really don't want to go down the path where we continue with bumpmaps and such, to simulate geometry, and then try to fix it with SSAA. What you're fixing is geometry aliasing, because the geometry is faked.
Bumpmapping in itself is flawed already, so why pursue it?
The future is obviously in more detailed geometry, and tessellation is one way to make that more efficient (which Pixar also uses). And when you use real geometry, you no longer need SSAA, because MSAA will deliver the same results.
Ofcourse one can also argue that as soon as you go down to micropolygon level (< 1 pixel, like Pixar's REYES rendering technique), the difference between MSAA and SSAA disappears completely. However, this obviously only goes for very detailed geometry.
Since we're doing per-pixel shading anyway, there's no reason to use high tessellation for (nearly) flat surfaces... or should I say: low frequency geometry. So that's where MSAA will still give you better efficiency.
 

BFG10K

Lifer
Aug 14, 2000
22,709
2,995
126
Yes, but since gradients and texture lookups are all filtered, these generally don't cause shader aliasing. It's really mainly just normal mapping and related effects that simulate geometry, which introduce sampling issues.
Any shader that is being undersampled will show aliasing.

Filtering is already applied for everything that just follows the surface contours.
AF tackles a different class of problem to SSAA. AF employs a non-square sampling pattern on texture surfaces to reduce their distortion without introducing more aliasing.

Why do you think texture filtering exists?
See above. Why do you think TrSS/AAA exists?

In fact, SSAA is not exactly a good way to battle texture/colour aliasing issues.
Actually it’s the best way. Even perfectly rendered 512xAF can show visible texture aliasing without super-sampling. Any tester app or filtering whitepaper will readily demonstrate this.

You need a LOT of supersampling to handle things that a normal anisotropic texture filter can handle in a single shader lookup.
Nope, even 2xRGSS provides a huge gain in image quality, as can be readily demonstrated in hundreds of games with screenshots and in-game movement. Anybody that has played modern games and tried SSAA can confirm this.

Anyway, nobody is saying SSAA is replacing AF. What’s being stated is that AF isn’t good enough to cure texture and shader aliasing by itself, so super-sampling is needed in conjunction with it.

People using SSAA to try and cure texture aliasing always gives me a bad taste in my mouth.
They aren’t trying, they are curing it. 2560x1600 with 8xSGSS will provide perfect image quality with zero aliasing in most cases.

I don't think texture filtering will ever be replaced by screen-space supersampling, because they just solve different problems.
Again, nobody ever stated otherwise. We’re simply stating that you need both because AF by itself is not good enough.
 

Scali

Banned
Dec 3, 2004
2,495
0
0
Any shader that is being undersampled will show aliasing.

Sorry, I guess I went too quickly for you...
I was already at the next step: What makes it that shaders can be undersampled?

The rest is just completely beside the point (and a bunch of nonsense as well, SSAA to cure texture aliasing instead of texture filtering? You STILL had to insist on that? AAAAAAARGHHHHH#@*()*@!#(*@!). I don't think you understood my earlier posts (see also what I said to SirPauly), so I'll just ask you to read it again, because my point still stands.
 

BFG10K

Lifer
Aug 14, 2000
22,709
2,995
126
What makes it that shaders can be undersampled?
I would've thought this was quite obvious: not enough pixels to sample with.

The rest is just completely beside the point (and a bunch of nonsense as well, SSAA to cure texture aliasing instead of texture filtering? You STILL had to insist on that? AAAAAAARGHHHHH#@*()*@!#(*@!).
For the third time, nobody is stating SSAA replaces AF. What is being stated is that AF by itself isn&#8217;t enough, and hence needs to be used in conjunction with SSAA.

The images below show documented proof that AF by itself isn't enough and that SSAA reduces texture aliasing, unlike your claims to the contrary.

If you can&#8217;t see the massive reduction in texture aliasing in the bottom image, then I can&#8217;t explain it to you any simpler.

16xAF:

72176070.png

72176070.png


16xAF + 8xSGSS:

afss.png
 

Scali

Banned
Dec 3, 2004
2,495
0
0
I would've thought this was quite obvious: not enough pixels to sample with.

I wasn't asking the question, I was answering it:
We have filtering and AA systems that work fine for matching the geometry of polygons... but as soon as you start faking geometry inside a shader, the filtering and AA falls apart... With tessellation, you can use real geometry, and get correct filtering again (like in the days before normalmapping).

For the third time, nobody is stating SSAA replaces AF. What is being stated is that AF by itself isn&#8217;t enough, and hence needs to be used in conjunction with SSAA.

I *was* stating that some people think that SSAA replaces AF.
What you're bringing up is a completely unrelated point. SSAA will 'blur' texture filtering issues, but that does not necessarily make the results correct.
But that's not the point here. The point here is that the shimmering-type effects that you see in modern games like Crysis comes from faking geometry in a pixelshader, where texturefiltering does NOTHING to avoid this aliasing... since MSAA basically means that supersampling is not applied in the middle of polygons, because of the assumption being made: inside the polygon, all gradients are linear, and linear filters are in place already.
This ofcourse is no longer true when you use a normalmap, because it will break the linearity of the polygon surface. You get aliasing because you break the assumption under which MSAA does its work. Tessellation will not do that. It will add more polygons instead, and each polygon retains its linearity.
 
Last edited:

evolucion8

Platinum Member
Jun 17, 2005
2,867
3
81
I would've thought this was quite obvious: not enough pixels to sample with.


For the third time, nobody is stating SSAA replaces AF. What is being stated is that AF by itself isn’t enough, and hence needs to be used in conjunction with SSAA.

The images below show documented proof that AF by itself isn't enough and that SSAA reduces texture aliasing, unlike your claims to the contrary.

If you can’t see the massive reduction in texture aliasing in the bottom image, then I can’t explain it to you any simpler.

16xAF:

72176070.png

72176070.png


16xAF + 8xSGSS:

afss.png

Those are results from an nVidia card right? I can smell it from a thousand miles of distance. :)

I do think that super sampling alone will not help in reduce shader aliasing, specially with its blurrying effect (Metro 2033 uses similar technique when 4x FSAA is used and looks bland compared to AAA which looks way more sharp).
 

BFG10K

Lifer
Aug 14, 2000
22,709
2,995
126
We have filtering and AA systems that work fine for matching the geometry of polygons... but as soon as you start faking geometry inside a shader, the filtering and AA falls apart... With tessellation, you can use real geometry, and get correct filtering again (like in the days before normalmapping).
No, they do not work fine. Any shader or texture can show aliasing if it’s undersampled, and your example is but a small subset of this problem. Even today’s polygon edges with MSAA applied to them still look jagged and not completely smooth because of the shader aliasing not being taken care of.

Also games well before normal mapping still suffered from texture and lighting aliasing, and AF was not enough to clean it up. I can list several games from 1999 that show visible benefits from super-sampling that I’ve actually witnessed in-game.

I *was* stating that some people think that SSAA replaces AF.
Where? Not in this thread that I’ve seen, so you’re arguing a strawman.

What you're bringing up is a completely unrelated point. SSAA will 'blur' texture filtering issues, but that does not necessarily make the results correct.
All forms of AA are blurring, namely the removal of high frequencies that couldn’t be displayed properly to begin with. And actually some forms of SSAA actually sharpen the image because they stack with AF, and I have documented evidence of this happening.

Tessellation will not do that. It will add more polygons instead, and each polygon retains its linearity.
Again, those polygons still need to be shaded and textured internally, and anything undersampled in there will show aliasing.

Did you understand the pictures I posted? Do you acknowledge the visible texture aliasing that 16xAF still has, despite not a single shader being around? Do you acknowledge the huge reduction of texture aliasing that SSAA delivers?
 

BFG10K

Lifer
Aug 14, 2000
22,709
2,995
126
Those are results from an nVidia card right?
Yep, and the Radeon’s 16xAF looks even worse. :p

I can smell it from a thousand miles of distance.
Absolutely, but I guess some people like to argue even when evidence that proves them wrong is right in front of them.

I do think that super sampling alone will not help in reduce shader aliasing, specially with its blurrying effect (Metro 2033 uses similar technique when 4x FSAA is used and looks bland compared to AAA which looks way more sharp).
There seems to be some confusion here. The 4xAA setting in Metro is actually supposed to be MSAA, but there’s some kind of problem with its implementation. Properly implemented MSAA (and SSAA for that matter) doesn’t blur things like that.

As for the game’s AAA, that’s actually a form of SSAA, and ironically it looks far better than the game’s “MSAA”.
 

brybir

Senior member
Jun 18, 2009
241
0
0
Absolutely, but I guess some people like to argue even when evidence that proves them wrong is right in front of them.


Welcome to the Video Cards and Graphics Forums on Anandtech, you must be new here!
 

Scali

Banned
Dec 3, 2004
2,495
0
0
No, they do not work fine. Any shader or texture can show aliasing if it&#8217;s undersampled, and your example is but a small subset of this problem. Even today&#8217;s polygon edges with MSAA applied to them still look jagged and not completely smooth because of the shader aliasing not being taken care of.

Also games well before normal mapping still suffered from texture and lighting aliasing, and AF was not enough to clean it up. I can list several games from 1999 that show visible benefits from super-sampling that I&#8217;ve actually witnessed in-game.

Problem is, you're blaming it on MSAA and offering SSAA as a solution.
This has two problems:
1) The problem is not caused by MSAA, but by less-than-perfect implementations of AF and mipmapping. Therefore the solution is better AF, not bruteforce AA.
2) SSAA isn't the correct way to handle poor AF output. It may look smooth, but so does a blur filter over the entire scene. So you may as well use MSAA and let post-processing deal with the blurring (tone-mapping, DOF etc).

All forms of AA are blurring, namely the removal of high frequencies that couldn&#8217;t be displayed properly to begin with.

The problem is when you are creating high frequencies because of broken sampling. Your filter is actually *introducing* aliasing, rather than removing it.
You're just bruteforcing your way through a broken image.
(What's the point of supersampling pixels on a linear surface? You already know exactly where your neighbouring pixels are, so you can implement that knowledge easily in the texel fetch itself. That's the whole point of using triangles rather than any other kind of primitive, linear surfaces are easy to interpolate and filter, because you can exploit the locality).

Again, those polygons still need to be shaded and textured internally, and anything undersampled in there will show aliasing.

But linear surfaces (polygons) will not cause shading issues (an interpolated surface normal is a 'perfect' normal at each position, it cannot be undersampled or oversampled, you have a perfect mathematical solution of 'infinite' precision). They would also not cause texture sampling issues when using a proper texture filter.

Which brings us back to the old point of "you can only spend your cycles once".
A proper MSAA+AF solution can get VERY close to an SSAA solution when you only use linear surfaces. So close that the huge extra investment of GPU cycles required for SSAA is not a good tradeoff. Those cycles are much better applied to other areas. Higher tessellation, more realistic shading, better post-processing etc.
 
Last edited:

evolucion8

Platinum Member
Jun 17, 2005
2,867
3
81
Yep, and the Radeon’s 16xAF looks even worse. :p

Yeah, and its odd considering that AMD's AF is totally angle independant, and yet, the shimmering and some moire artifacts are still noticeable.


Absolutely, but I guess some people like to argue even when evidence that proves them wrong is right in front of them.

Yeah, typical of humans. :p

There seems to be some confusion here. The 4xAA setting in Metro is actually supposed to be MSAA, but there’s some kind of problem with its implementation. Properly implemented MSAA (and SSAA for that matter) doesn’t blur things like that.

As for the game’s AAA, that’s actually a form of SSAA, and ironically it looks far better than the game’s “MSAA”.

Yeah, I had the same experience with that, it looks like supersampling and the performance impact is steep, but I hear that's related to AMD hardware but I have seen some nVidia users reporting the same issue.
 

BFG10K

Lifer
Aug 14, 2000
22,709
2,995
126
1) The problem is not caused by MSAA, but by less-than-perfect implementations of AF and mipmapping. Therefore the solution is better AF, not bruteforce AA.
Yes, the problem is with AF, and there’s no real-time implementation that will resolve all texture and shader aliasing. Even 512x ALU-rendered AF shows similar problems, and that’s not available outside of tester apps and similar. Therefore the only available solution is to use SSAA which visibly reduces shader and texture aliasing.

2) SSAA isn't the correct way to handle poor AF output. It may look smooth, but so does a blur filter over the entire scene. So you may as well use MSAA and let post-processing deal with the blurring (tone-mapping, DOF etc).
Again, properly implemented SSAA shows little to no blurring, and in fact some forms actually sharpen the image because they stack with AF. This is unlike a post-filter which is typically designed to blur things and doesn’t stack with AF.

But linear surfaces (polygons) will not cause shading issues (an interpolated surface normal is a 'perfect' normal at each position, it cannot be undersampled or oversampled, you have a perfect mathematical solution of 'infinite' precision). They would also not cause texture sampling issues when using a proper texture filter.
This is simply untrue; any surface can be undersampled if there are not enough pixels. That’s a by-product of converting from world space to screen space.

A proper MSAA+AF solution can get VERY close to an SSAA solution when you only use linear surfaces.
Not in games with shader and/or texture aliasing it can’t. In such games even polygon edges aren’t perfectly smooth because the MSAA can’t deal with the shader and/or texture aliasing. This problem is very visible in many games from 2004 onwards.

So close that the huge extra investment of GPU cycles required for SSAA is not a good tradeoff. Those cycles are much better applied to other areas. Higher tessellation, more realistic shading, better post-processing etc.
I’m starting to wonder whether you have any significant gaming experience at all. Even 2xRGSS makes a huge difference in many games from 2004 onwards, and anybody that has tried it knows this.

Also elementary benchmarking reveals the performance is usually between 4xMSAA and 8xMSAA while offering much better quality overall in aforementioned games.
 

BFG10K

Lifer
Aug 14, 2000
22,709
2,995
126
Yeah, and its odd considering that AMD's AF is totally angle independant, and yet, the shimmering and some moire artifacts are still noticeable.
ATi undersamples the surfaces compared to nVidia. Not only do they have more texture aliasing, but they also have sharp filtering bands which are very visible in-game during movement.

Yeah, I had the same experience with that, it looks like supersampling and the performance impact is steep, but I hear that's related to AMD hardware but I have seen some nVidia users reporting the same issue.
The same problem happens on nVidia hardware too. It&#8217;s a game issue related to its &#8220;MSAA&#8221; implementation, which is actually broken and isn't really MSAA at all.
 
Last edited: