Do 'retina' displays eliminate the need for AA?

[tynopik]

Would a game running on a 'retina' display at native resolution receive no benefit from AA?

 

[Kloreep]

I'm no expert, but I would imagine there'd still be some benefit.

As I understand it, the basic problem that AA is there to correct, is that engines generally render each pixel as occupied by one object (or side of an object) only. Hence frequent aliasing or other problems being apparent with objects that have fine and noticeable lines to them, like railings or wire fences. AA's ability to blend edges and objects together helps remove or reduce seams and make everything look more coherent.

I'm sure "retina" displays would reduce the benefit of AA, just as larger resolutions generally do. (Or more accurately: AA-lessness might be less noticeable.) But I don't think they would remove it.

 

[ViRGE]

Yes and no.

Keep in mind that aliasing is the result of generating images of (and mapping them to) a display with discrete elements (pixels), and then shown to beings with continuous vision. The ultimate solution to aliasing would be to have a display with no discrete elements, effectively giving it an infinite resolution.

To that end a retina display cannot solve aliasing, but it can reduce the impact. With higher pixel densities the individual pixels are smaller, and as a result the aliasing is less noticeable. It's still there, but smaller pixels will be harder to notice, and having additional pixels in the same amount of space allows for more continuous transitions (e.g. specular lighting). Depending on the distance 300dpi may be good enough, but it's more likely that you're going to need higher pixel densities than that to really mask all forms of aliasing.

 

[Concillian]

I think there will be some form of AA needed even with high ppi monitors. The way game engines work by layering textures on textures results in funky transitions when something like this happens at high contrast. Our eyes just don't see that kind of thing in real life where objects that are supposed to be 40 yards away are so crisp and detailed. In many respects, games are 'too accurate' in distance and partially transparent cases and these need some work. Lighting is part of it, but AA plays a role too.

I strongly believe that we will move to better and better forms of post processed AA, and that will be more compatible with truly high resolution displays 200+ pixels per inch. since the horsepower required will increase with the resolution required to have 200 ppi on a ~20-27" screen.

Ultimately it will be a combination of game engine design, programming, and higher resolution as we move forward. As people are often quick to point out, a regular DVD at 640x480 is pretty detailed. There is a lot of room to grow in just getting better rendering design, I think that will go hand in hand with true resolution improvement.

 

[Soccerman06]

Still waiting on my 8k 30" monitor before I complain about AA :hmm:

 

[Bateluer]

A retina display is a marketing term for a high resolution display. AA gets more of a benefit as the resolution is increased.

 

[Lepton87]

A retina display is a marketing term for a high resolution display. AA gets more of a benefit as the resolution is increased.

My understanding of a retina display is that pixel density is so high that a human eye can no longer see individual pixels. Can someone calculate what resolution would a 24" screen have with a PPI of 320?

 

[lavaheadache]

My understanding of a retina display is that pixel density is so high that a human eye can no longer see individual pixels. Can someone calculate what resolution would a 24" screen have with a PPI of 320?

I'm sure you can do the simple math to figure it out.

 

[lavaheadache]

A retina display is a marketing term for a high resolution display. AA gets more of a benefit as the resolution is increased.

AA will show less benefit as resolution is increased as long as size stays the same

 

[KompuKare]

Can someone calculate what resolution would a 24" screen have with a PPI of 320?

Well wiki lists that a 24" 16:9 screen measures 53cm by 30cm, which is 20.87" by 11.81". At 320PPI that would be a resolution of 6677 by 3780 although if such a monitor were to come into existence I'm sure the actual resolution would be rounded to something like 6800 x 3800 or so.

 

[kidsafe]

We're basically talking something along the lines of 6528x4080 for 16x10.

 

[Lepton87]

We're basically talking something along the lines of 6528x4080 for 16x10.

Those would be true 4k panels and not what marketers decided to call 4k.

 

[ViRGE]

Well wiki lists that a 24" 16:9 screen measures 53cm by 30cm, which is 20.87" by 11.81". At 320PPI that would be a resolution of 6677 by 3780 although if such a monitor were to come into existence I'm sure the actual resolution would be rounded to something like 6800 x 3800 or so.

Keep in mind that for compatibility reasons it's likely that any HiDPI monitors will be integral multiples of existing monitors. So a HiDPI 24" 16:9 display would need to be 7680x4320; 16 times the resolution of a 1920x1080 monitor.

 

[soccerballtux]

I'm sure you can do the simple math to figure it out.

that's not very sensitive. What if he's stupid, or worse, lazy? Now look what you've done.

 

[Arkadrel]

Would a game running on a 'retina' display at native resolution receive no benefit from AA?

The resolutions needed, to make AA unnessary, are to demanding hardware wise, to be a solution that bests AA.

Otherwise we would all be useing 4k or better moniter resolutions already....

 

[Anarchist420]

1920x1200 front buffer resolution is plenty. Lossless textures and higher precision back buffers are more important IMO.

Another thing is that if we have 4k x 2.5k front buffers, then the back buffers would need to be that high. That would be a waste of graphics memory. I guess we're just going to see more lossy compression and/or lower back buffer precision/no or less AA since all so many people care about is resolution. I'm not suggesting that we go back to 800x600 resolution, but increasing front buffer resolution requires an increase in back buffer resolution (because even the best interpolation doesn't cut it) and increases in back buffer resolution tend to increase memory requirements more than any thing else.

 

[lavaheadache]

Who cares if it increases memory requirements? Put more memory on cards and increase bandwidth. It is called progression for a reason. I'm not saying blow efficiency out of the water but surely an increase in resolution should warrant higher memory constraints and related things.

Some comments make me wonder how many people would still be content with DOS as an OS and Dot matrix printers. While we are at it I think even 56k modems are too fast for what we need.

 

[soccerballtux]

Who cares if it increases memory requirements? Put more memory on cards and increase bandwidth. It is called progression for a reason. I'm not saying blow efficiency out of the water but surely an increase in resolution should warrant higher memory constraints and related things.

it's mainly the memory bandwidth that is the killer. Larger buses require more pins, and cramming more pins onto the dies is hard, and routing traces for all those BGAs is where the real expense kicks in. That's why it's economically feasible for GPU makers to be 2+ generations ahead of desktop memories as far as DDR generation goes.

 

[BrightCandle]

A true retina display (one where the eye can not tell the difference if more pixels were present) would eliminate the need for AA. Today's so called retina screens however are not quite that good, its mostly marketing for a screen designed to be used closer than a Monitor/TV. The inability to see the pixels is not the same as not being able to tell the difference, so we have a way to go before AA is eliminated.

 

[Ares1214]

We're basically talking something along the lines of 6528x4080 for 16x10.

I'm hoping we find a more elegant and advanced way of displaying images by the time resolutions reach that point. A breakthrough in the way we display things would likely be spurred on by a breakthrough in GPU's, so basically semiconductors all together. You will never see that resolution the way things are going now, since we have been stuck at "HD" for years now, and a resolution increase of that size would be 13x the amount of pixels, and you would need 13x amount of the GPU power to display that.

 

[bryanW1995]

Who cares if it increases memory requirements? Put more memory on cards and increase bandwidth. It is called progression for a reason. I'm not saying blow efficiency out of the water but surely an increase in resolution should warrant higher memory constraints and related things.

Some comments make me wonder how many people would still be content with DOS as an OS and Dot matrix printers. While we are at it I think even 56k modems are too fast for what we need.

Sigh...I remember when I first "high speed" internet at my house. The porn, uh, I mean, "work-related high bandwidth photos" loaded so much faster...ah, the memories.

Seriously, we've been stuck at 2560x1600 for far too long. Let's at least get some 2k monitors out there!

 

[Soccerman06]

Those would be true 4k panels and not what marketers decided to call 4k.

Wrong and kinda right, 4k refers to a panel of 4000 or more horizontal pixels (usually 4096x 2160). The market which your referring to QHD as 4k which is 3840x2160.

 

[lavaheadache]

Honestly, I would be happy if they made 2560x1600 24 " panels.

 

[taltamir]

Keep in mind that aliasing is the result of generating images of (and mapping them to) a display with discrete elements (pixels), and then shown to beings with continuous vision.

Humans do not have a continuous vision. They have a discreet countable amount of cones and rods in (an imperfect) pattern.

Would a game running on a 'retina' display at native resolution receive no benefit from AA?

AA is an optical illusion created by blending nearby pixels to mask the fact an image is rendered unto pixels which are large enough for a human to distinguish between them.

Being projected unto a retina is not a magic bullet for solving this issue. Although it does make things simpler in some regards (and more difficult in others, see below)
Ultimately the resolution/dot pitch is what matters most for your ability to detect jaggedness.

The human retina contains about 120 million rod cells (color blind, sensitive to low light and motion, primarily peripheral vision) and 5 million cone cells (3 types for 3 colors, require bright light, centralized at middle of eye).
Due to the brain processing the data organically it probably does NOT consider each receptor cell to be a pixel and thus the distinguishable resolution is somewhat lower. I'm not quite sure if it uses rods to augment resolution but if it does then not by much considering how poor peripheral vision is.

However http://hyperphysics.phy-astr.gsu.edu/hbase/vision/rodcone.html
As you can see cones are extremely dense in the center of the eye and thus you do not simply have a color resolution of 5 megapixels, but a focus resolution of 5 megapixel (or rather, probably somewhat under that due to the organic nature of the brain). Rod cells are color-blind and make up primarily (but not exclusively) the peripheral vision.

At some point the resolution would be high enough that the human eye could not distinguish jaggedness and AA would not be needed at all. Whether it is projected on the retina or from a monitor.

I would say then that a display should be 5 megapixels per area of focus.
So you need to measure a human's focus area, compare it to distance they sit from display, and make it have enough pixels so that no matter where on the display you focus you cannot distinguish jaggedness.
A retinal projection would have the advantage of being high resolution central display focused directly unto the center of the eye and secondary peripheral low resolution display for the peripheral vision. This would greatly reduce the total amount of pixels needed to eliminate the need for AA.

Some maths.
If a human focus area (at the distance you sit from a monitor; this varies and can be better controlled with retinal projection display) is 10cm diameter circle (area = 25cm*pi) and can distinguish 4 megapixels (hypothetical numbers, I am not sure if true) then it has a pixel density of 0.0509 megapixels/cm^2
A 16:10 display that is 24 inches diagonally is 20.35 cm by 12.72 cm aka 258.88 cm^2. And would in the above scenario need to have need to have 13.18 megapixels for them to correspond 1 to 1 to cells in the eye and be indistinguishable such that AA would be obsolete.
4K displays are almost at that, but since I just guessed at the focal area and resolution earlier on there is no way of telling without proper research to actually deduce the correct figures.

 

[lavaheadache]

nice explaination, however I'm fairly certain he was refering to the term "retina display" that apple coined for the iphone 4