🎉 Celebrating 25 Years of GameDev.net! 🎉

Not many can claim 25 years on the Internet! Join us in celebrating this milestone. Learn more about our history, and thank you for being a part of our community!

Is this type of RTS possible? (polygon count)

Started by
47 comments, last by Newgamemodder 3 years, 2 months ago

Newgamemodder said:
Does anybody know if there is less performance use if the characters are organized into “squads” of 9 characters in each?

Yes, every logic you add costs something.

Advertisement

10-12 billions seems like it would be outside what computers could manage. No CPU would be fast enough to process that many objects. Even high end CPUs these days only clock out at around 4ghz. That would be around ⅓ a clock cycle per unit. Even if you threw 16 cores at it there would only be about 5 cycles per unit so you really couldn't do much.

You would also use a TON of memory. Suppose you only stored the minimum for a unit. An X/Y coordinate as a 16bit fixed point number. That is still 4 bytes per unit and would take about 40gb of RAM and 16 bits would likely not be enough.

To pull something like this off you would have to make a game engine specific to this task. Instead of simulating individual units, you would have to simulate groups of units which would probably work just fine since the player couldn't possibly comprehend billions anyway. If I were approaching this I would break the game world into blocks. Each block would be capable of handling 100s of thousands or even millions of units each. So instead of listing out each unit individually a block simply specifies how many of each unit type is contained in that block. You could then run a high level simulation of what happens in that cell based on the number of units in it. You could also do some rendering shortcuts where you draw large batches of enemies at once probably representing dozens, if not hundreds, of enemies per textured triangle.

Depending on your viewpoint in the game you could implement a level of detail engine in both the graphics and simulation. Once you zoom into a cell far enough you would start rendering and simulating individual units. One challenge you would have to solve is some form of consistency when moving back to a cell you already have seen. You would also need some way to prevent the player from trying to put all of their units in one place ruining the level of detail system.

Basically you would need to make a custom engine to tackle this problem and would have make some compromises to make it work. Depending on what sort of gameplay you had in mind, you may be able to make something like this work.

My current game project Platform RPG

Just want to make sure i've understood correctly,

I multiply the amount of triangles on scene x fps correct?

I can probably halv the number of units ingame together with polygon count. that would be about 1840 units on screen, minus turrets, buildings and props.

Is up to 8 billion still to much?

(about 1.8 billion is for props, walls and buildings). So any ideas how many polygons props are in rts? also are RTS games 30 fps or 60 fps?

The only reason the polgon count is so massive is because it's a scanned 3D model.

I can probably lower the polygon count of the infantry

Ok so without dead bodies my new polygon count is 8 billion tris (without optimising infantry/getting a more realistic polygon count of props/foliage, buildings and walls)

About 1800 soldiers (200 squads of 9 soldiers in), 40 vehicles, 200 7k tris soldiers in 40 squads of 5 each) ALSO this is the ABSOLUTE WORST SCENARIO so i doubt a battle may look like this.

Why would your soldiers be 7k triangles? In an RTS game where you camera is far away, you probably can perceive between 2 and 100 triangles depending how far you are. You would have to zoom all the way on a single soldier to have it fill your entire screen to be able to see the 7k triangles.

That's the highest lod, MIGHT be able to make them less

Newgamemodder said:
I multiply the amount of triangles on scene x fps correct?

To me it's just confusing to multiply with fps. I then need to to divide (by how many fps?) to make some sense out of it.

It may be GPU vendors tell us such numbers like ‘million triangles per second’, but that's just theoretical max throughput, and i still don't know how large those ‘triangles’ are. I ignore such numbers because not useful in practice. Maybe for marketing.

Newgamemodder said:
The only reason the polgon count is so massive is because it's a scanned 3D model.

Huh? So, you mod a scanned model into some RTS game? As background terrain, or what?

There are tools to decimate triangle count without visual loss. In general, scanned data needs a lot of post processing to become efficient for games.

JoeJ said:

Newgamemodder said:
I multiply the amount of triangles on scene x fps correct?

To me it's just confusing to multiply with fps. I then need to to divide (by how many fps?) to make some sense out of it.

It may be GPU vendors tell us such numbers like ‘million triangles per second’, but that's just theoretical max throughput, and i still don't know how large those ‘triangles’ are. I ignore such numbers because not useful in practice. Maybe for marketing.

Newgamemodder said:
The only reason the polgon count is so massive is because it's a scanned 3D model.

Huh? So, you mod a scanned model into some RTS game? As background terrain, or what?

There are tools to decimate triangle count without visual loss. In general, scanned data needs a lot of post processing to become efficient for games.

I read somewhere i had to multiply by fps. So i don't need to?, in that case it's 118/119 million polygons on screen at once.

The scanned mesh is from another game. It's a character. Is it still possible to decimate it more while keeping the detail level/curvateur?

Newgamemodder said:
I read somewhere i had to multiply by fps. So i don't need to?, in that case it's 118/119 million polygons on screen at once.

I would say, when communticationg polygon counts, you usually tell a number your asset(s) have. No need to involve time here, IMO.

Still, i have to repeat my first response: A screen at 1800 p has about 2 million pixels. 4k so has about 8. Thus, it's complete nonsense to display 120 millions on such screen. You only amplify aliasing because each pixel snaps to a somewhat random triangle out of many. It would look like noise and bad.

Newgamemodder said:
The scanned mesh is from another game. It's a character. Is it still possible to decimate it more while keeping the detail level/curvateur?

How many triangles does the mesh have? Was it meant for cutscenes, maybe, and now you want to scale it down to RTS unit size of 30 pixels high or less? No good idea. You really have to reduce it. And yes, you won't notice a difference in quality from the reduction.
It's however a difficult task. For example, you may need to preserve UV chart boundaries so textures still work, and the same about different materials. Also, game meshes are cut along such boundaries, which makes the reduction much harder, not having the original data. You also want to preserve bone influences for skinning and such things. I assume Blender has the necessary tools, but it's not easy and requires some practice / research on the workflows.

I've found some games that use high polygon characters in strategy games. Since i frequent modding forums i know polygon counts of meshes.

Each one of these clone troopers are about 18-22k tris (without weapons)

These if i recall can be anything between 8K and 20K tris

Well the character with the highest polygon count for my mod is 44k +10k for weapon. The lowest is probably 34k. It's armor so keeping curvateur is the most important. Can one keep the curvateur if one lowers the polygon count to 15k tris?, or can i lower it more but keep detailing/curvateur?

Can one keep the curvateur if one lowers the polygon count to 15k tris?

Depends on the tool in use, but preserving curvature is usually the main objective of such optimization algorithms. (Beside UV / material borders).

On the other hand, character model lods are usually made manually, because results of automated processes are still somewhat meh.

44k sounds pretty average for FPS. Top down RTS ofc. surely use less than that. 44k would only make sense when zooming in. I see making manual lods for characters is too much work for your purpose, but you can always give it a try.

Btw, Simplygon is industry standard and has a free version with online processing. Maybe it does the trick with little effort after some trial and error.

This topic is closed to new replies.

Advertisement