Do game devs/LOD systems try to do 1 triangle per pixel?
Usually the target is ‘not smaller than 10 pixels’.
Otherwise GPUs are poorly utilized, which is why Epic made a software rasterizer in compute, doing >80% of triangles in their demos.
Newgamemodder said:
If thats the case 16K resolution
Won't happen. We already address 4K with upscaling.
Moores Law is dead. GPUs are too expensive. Current gen is next gen. XD
Usually the target is ‘not smaller than 10 pixels’.
So my calculation of 132-137 in the camera frustum is impossible?
“Won't happen. We already address 4K with upscaling.”
Do you mean that gaming wont go beyond 8K?, or that we already have 16K via upscaling 4K?
Newgamemodder said:
or that we already have 16K via upscaling 4K?
Or equally likely, by upscaling 1080p/2440p even further. A lot of current gen games already dynamically vary the render resolution based on moment-to-moment performance, and with tech like DLSS, render resolution becomes even less strongly related to display resolution.
Newgamemodder said:
Do you mean that gaming wont go beyond 8K?, or that we already have 16K via upscaling 4K?
I mean it won't go there, eventually. It makes sense only with huge screens. And we have other issues like low brightness displays. Improving this makes more sense than even higher resolutions.
Performance wise, such high resolutions would become possible withe gaze tracking and foveated rendering. I don't believe temporal subpixel upscaling (such as DLSS as well) could upscale 4x good enough.
Finally, increasing use of raytracing also speaks against higher resolutions.
ok but how does it affect what i am striving for?, or is my entire idea impossible?
The only reason i need 16K is the polygon count on screen :/
When it comes to 4K i was told “what is the point in having so high polygon count when you can only see 9 mil”, what did he mean by that?
Newgamemodder said:
When it comes to 4K i was told “what is the point in having so high polygon count when you can only see 9 mil”, what did he mean by that?
Well, as said: Having more triangles than pixels means more detail than you can show, but wasted power and increased aliasing.
Newgamemodder said:
The only reason i need 16K is the polygon count on screen :/
If you zoom in, the detail shows even at 1080p. That's why FPS games have the highest texel density for textures. A TPS game already needs less because camera does not come so close to surfaces.
A strategy game which never zooms can set it's geometry / texture resolutions to the constant scale that is shown on screen.
Newgamemodder said:
ok but how does it affect what i am striving for?, or is my entire idea impossible?
If it's possible or not depends on engine and hardware. As discussed already.
If it makes any sense is the question you should think about first. Is the super high detail you want to have a visual improvement over the lesser one? What additional details would we see?
Thinking of Star Wars, I remember surface details on space ships: Small features of boxes and plates, some cylinders, etc. It's a decoration of the overall shape which is defined at much lower frequency.
The question then is: Do we need actual geometry to show those patterns, or is a texture enough? Is there any other small scale detail which has a function or artistic expression, like turrets, doors, etc., which is worth to be seen precisely?
Or are all those tiny details just uniform high frequency noise, which means there is no real reason to show them?
So i've changed the polycount to 6.9 mil per ship
The scene in total is now 132228000/132710400
So i've optimized it for 16K. Now what do i do about resolution?, since you said 16K is unlikely :/ Do i have to count DLSS or something else?
Hi again!
so i managed to fit my polyogn count within the pixelcount of 16K. There is about 160k left for props.
I'm sorry but i still don't understand the “Well, as said: Having more triangles than pixels means more detail than you can show, but wasted power and increased aliasing.”
What happens if i have say 3-5 million above the pixel count? (counting big ships, fighters and everything else)
A. Do polygons disapeer from the big ships?
B. Do i have problems seeing the details because there are more small triangles?
"If you zoom in, the detail shows even at 1080p. That's why FPS games have the highest texel density for textures. A TPS game already needs less because camera does not come so close to surfaces.
A strategy game which never zooms can set it's geometry / texture resolutions to the constant scale that is shown on screen."
So i don't need 16K? :D
the only reaosn i want 16K is because i want the highest detail zoomed out in a fleet like this:
I could have less for the ships but i've recieved so many different answers to what polygon count this ship can be made with :/
A. Do polygons disapeer from the big ships?
B. Do i have problems seeing the details because there are more small triangles?
If you have smaller triangles than a pixel, so multiple triangles and up in the pixels area, we have ‘subpixel detail’.
Currently that's not really supported. Standard rasterization selects the single triangle which is at the center of the pixel. Becasue there are many triangles, we so have the problem of ‘undersampling’, which can cause aliasing.
AA methods like TAA or MSAA take multiple samples at different subpixel locations and try to calculate an average of them, which then would come from multiple triangles. But because we can not see subpixel detail anyways, the practical goal here is more smoothing triangle edges, or flickering specular highlights, etc.
So it's clear that having triangles smaller than a pixel makes no sense visually, and more likely harms image quality than helping it.
Thus the goal of a perfect LOD solution on infinitely fast HW would be still ‘triangles not smaller than a pixel’.
Which does NOT mean you have to use models at one triangle per pixel density. Less detail may still be good enough, more efficent, less storage, and even more future proof.
Newgamemodder said:
So i don't need 16K? :D
Even if 16K screens become widespread in the future, it does not mean we need to increase details of content once more again. Each time we double resolution, the noticeable improvement becomes smaller.
On a 16K screen you could still see individual pixels, but only if there is high contrast and aliasing. A good image will not look so much more detailed than on a 4K screen.
Remember the big improvement s from SD to HD, but then the much less noticeable improvement from HD to 4K.
The next step just makes little sense if we trust our eyes instead marketing promises, and considering world economy appears a bit pessimistic eventually, people might just not buy in to that ‘resolution and upscaling’ craze anymore, imo.
I would not plan or think about 16K on your side, but it's up to you.
JoeJ said:
Well, as said: Having more triangles than pixels means more detail than you can show, but wasted power and increased aliasing.
So technically i can have 200 mil at 4K or 1080p?, if so how are the details affected?
Sorry if i seem slow but i still don't understand this part :/
Sadly i don't know if my project is possible anymore because i have heard reports i need my 3D mesh to e 10 million to get all the details :(
