79 replies to this topic

[Dmm]

Also AA is a real killer in 4k.
 

Hopefully one won't need AA with a 4k monitor.

[Lee-m]

Hopefully one won't need AA with a 4k monitor.

You always need some AA regardless of the resolution. Hair for example is a big issue even on 4k, along with water and other fx. Plus you still have LOD systems in game engines to work with.
...well I guess you never 'need' it, but I always notice when its not on or low. If I had spent good money on a 4k system I wouldn't want to leave the AA turned off lol

[toremygg]

Yeah, probably some video editing and conversion, which can be pretty cpu intensive.

[Veen Friend]

The AA would be a must also in 4K when stopping up in the game just to see landscapes or art.

The AA is a key option to switch on and off when needed, and then it's very brilliant to use.

[Dmm]

According to one article I read once the pixels reach 300ppi/118ppcm on the diagonal AA will no longer be needed. The jagged lines are still there but are no longer perceptible to the human eye. Even if that is true it going to take one very powerful graphic card to run such a monitor and is probably quite a few years off.

[DiskJunky]

NVidea's Titan Z would probably be able to manage. 12Gb VRAM

[Ragnar]

NVidea's Titan Z would probably be able to manage. 12Gb VRAM

 

That'd be like killing an ant with a nuclear weapon…but hey, sure, if you can afford it!

[Consili]

That'd be like killing an ant with a nuclear weapon…but hey, sure, if you can afford it!

when going on a bug hunt, a nuke is the only way to be sure...

[mrKnask]

when going on a bug hunt, a nuke is the only way to be sure...

Well, that depends

http://indianapublic...radiation-wins/

[UPtimist]

Man, this thread full of AA talk is filling my head with so many horrible Alcoholics Anonymous and The Automobile Association jokes that I don't even know where to begin... (Like, in case of a crash, it's good to have AA)

[Vainamoinen]

...and I always have to remind myself that "AA" doesn't mean Chapters is a AAA game minus one A.

 

 

 

Omygod what if AA means "Anti Aprilizing"?!!?

[DiskJunky]

That'd be like killing an ant with a nuclear weapon…but hey, sure, if you can afford it!

You're not kidding. I was speccing up an Alienware PC and came across the graphics card. It's about €1,200 on its own. I think I'll pass and get a 4gb card instead - I'm only running at 1920x1080 anyway

[Stargazer]

Will it run on my new Macbook Pro (late 2013) in native (retina) resolution?

[DiskJunky]

I'd imagine so - I think that's what Ragnar is doing his own testing on (I stand to be corrected of course)

[Stargazer]

That's awesome!

 

I spent waaaay too much on both my Macbook Pro and my pledge for Dreamfall Chapters. So naturally, they belong together .

 

Already looking forward to enjoying it in full resolution glory!

[Ragnar]

Will it run on my new Macbook Pro (late 2013) in native (retina) resolution?

 

Works for me!

[Sinerem]

Mac version huh?  Neat! 

[kamikaze]

I'm switching to 4k next week. But I don't think I have the graphics power to back it. It's just to have sharp fonts while working.

 

Any way, I wanted to pitch into the AA discussion.

 

IMHO the best AA method is oversampling. Everything else is just a performance compromise.

 

Some time based anti aliasing should offer a lot of visual improvement for almost no cost, which is a VERY SWEET DEAL in my opinion. Especially for hi-res gaming. Of course that wouldn't work if the framerate drops too low (yay for G-Sync which reduces the severity of that significantly). But at a too low FPS aliasing is the least of my worries.

 

Actually I'd love adaptive rendering algorithms that just produce a stable 60 fps. Scaling an image to the output resolution should be pretty predictable. Let's say you want to render 8k, because you're on a 4k system and want to use oversampling (we are adaptive so we can crank up the maximum as far as we want to):

- You start by rendering 480x270 px

- Then you render 960x540 px by calculating the 3*(480x270) px in between

- Then you render 1920x1080 px by rendering the 3*(960x540) px in between

- Repeat until you have used up more than a quarter of the rendering time available (that's the point where you know you won't finish with the next step in time)

 

As you can see each step doubling the res takes three times the time all the previous steps did (instead of 4 times because you can build on the previous step). All in all it should take exactly the same time as rendering the full res in the first place. The advantage here is that you're guaranteed to have something to put on the screen when you need it. You always get the maximum quality you can get and reduce power consumption, because you have some idle time between frames due to being able to bail out of rendering early.

 

You can combine this with time based antialiasing, because it doesn't cost extra and will greatly improve the visual quality of low res output. With the guaranteed fps it would work especially well.

 

Here is the catch - that would only work with raytracing/raycasting engines. The usual raster rendering approach just draws every triangle into a buffer (the Z-Buffer is for making sure that the front-most pixel of overlapping triangles wins) so you wouldn't be able to build on the previous resolution you rendered. Thus you have 4 times instead of 3 times the rendering time to reach the previous res (but it's 4 times the previous step, not 4 times everything). You wouldn't have a fixed bail point, but you'd have to predict from the previous res if you can still do 4 times that much.

It would still guarantee the frame rate but you wouldn't be able to reach the same resolution you would if you rendered the resolution you can reach directly. Because all the smaller resolution images are thrown away and were basically a waste of time and energy.

The best approach here would probably be to just render the small 480x270 res and then use that to predict how big you can go and render that.

[khh]

That assumes that the work that go into rendering a scene always scales linearly with the amount of pixels. Does it?

[kamikaze]

That assumes that the work that go into rendering a scene always scales linearly with the amount of pixels. Does it?

For raytracing it does. For what video games do right now - it can. But it doesn't have to. It depends on how the shaders were written. And on whether you have enough video memory. If you need to move texture data over the bus you're screwed.

 

Generally everything that scales > O(x*y) (memory boundaries aside) will come back to bite you and should be ditched/rewritten.