Ballpark Triangle Figures for Strategy games?

NanakoAC

Hello lovely people, this is my first post here. I'm an aspiring developer with a variety of prior experience, but it's been about 6-8 years since i last did 3D art for a game, in some Civ IV mods. Since then i've been a little out of touch and mostly focusing on coding.

Currently, I am interested in making a strategy game in unreal engine. I won't bore you with too many details, but i'm planning to do most of the work myself, art included. I'm brushing the dust off my old talents and firing up the latest version of blender, and I've realised I don't really have targets for performance yet.

I'm interested in making unit models with around the same detail level as the Total War series,  examples:

https://steamcdn-a.akamaihd.net/steam/apps/534330/ss_776fd4ca8eb250a3a6ddc9f852dbf494da0c104e.1920x1080.jpg?t=1533741513

https://steamuserimages-a.akamaihd.net/ugc/200804297983305039/0792B46BE73C3739AE74C64B888D7CCB3D8E3D19/

I'm posting on polycount because I found this place while trying to google performance stats for the latest Total War Warhammer game, (along with a nice thread where various artists were posting their official work, it was lovely!)

I flatter myself thinking i can emulate that quality, but I'll do my best. I need some ideas on targets though. That is, polycounts, and texture sizes. I'm used to working within tight limits, and I need to set such limits for myself so i don't run away spending a million triangles on an exquisitely sculpted nose or somesuch. But these days i have no idea what modern computers are capable of

So my question is:

For an average human figure, in armor or clothing, viewed from a sort of isometric perspective and in the context of a strategy game developed for modern gaming PCs:

What's a good target for his polycount (in triangles) and resolution for textures, normal maps, glow, gloss, etc.

Once I have target values for humanoids, I can use that as a baseline for performance targets for other things (like guys on horseback)

Any posts and thoughts are appreciated

poopipe

This sort of question is honestly pointless - nobody can answer it for you. 

Design the unit without restricting yourself, place it as close to the camera as it'll ever get in game and start reducing geometry and texel density - stop reducing when it starts to become visible to the player and you have your answer.

Modern consoles can draw millions of triangles on screen without breaking a sweat - materials are the biggest cost

Mark Dygert

It really depends on a lot of factors.
The hardware the game is running on? Mobile? Console? PC?

The engine it's using and how well it handles optimization? Do you need to micro manage the Levels of Detail or does it handle it for you? The time it takes to create assets is also a huge factor.

The overall design of the game? How many units are on the screen at one time? How close will the camera get? Will the same models be used for in-game cut scenes or cinematics?

Do the designers have a rough idea how many units will be on the screen? Will one character represent hundreds like the Civilization games or does every character need to live and die on their own?

What are the actions they preform, how many animations are needed, how many joints in the rig? How long does it take to create a character and rig it up, how many do you need to have? 7 recycled units or a hundred unique characters?

Are they just low poly models with simple textures, or will there be complex materials that need to be derived from normal maps meaning there will be a high poly -> low poly workflow.

How important are other aspects of the game, like the environment, AI, particles, back end systems?  

Often the tri count is heavily dependent on testing and performance balancing as you go. What you trim back in one game to give another aspect more impact, won't be the same as another game.

If I had to take a wild stab in the dark I would start testing with quickly done stand in models that start at roughly 20k tris and see how things preformed, making sure the rest of the game had more or less equivalent stand-ins. Keeping in mind that there might be a large optimization push again later.

NanakoAC

poopipe said:

This sort of question is honestly pointless - nobody can answer it for you. 

Design the unit without restricting yourself, place it as close to the camera as it'll ever get in game and start reducing geometry and texel density - stop reducing when it starts to become visible to the player and you have your answer.

Modern consoles can draw millions of triangles on screen without breaking a sweat - materials are the biggest cost

I can understand this being less of an issue than I imagined, but I doubt we've yet entered a post-polygon era where performance concerns no longer exist.

I'm feeling a little aimless and trying to find concrete targets to use as a starting point

If people can't give subjective advice, then at least one part of what I'm asking has a definite answer. What kind of polycounts are used in Warhammer: Total War. Or any other recent strategy games as an example?

NanakoAC

Mark Dygert said:

It really depends on a lot of factors.
The hardware the game is running on? Mobile? Console? PC?

The engine it's using and how well it handles optimization? Do you need to micro manage the Levels of Detail or does it handle it for you? The time it takes to create assets is also a huge factor.

The overall design of the game? How many units are on the screen at one time? How close will the camera get? Will the same models be used for in-game cut scenes or cinematics?

Do the designers have a rough idea how many units will be on the screen? Will one character represent hundreds like the Civilization games or does every character need to live and die on their own?

What are the actions they preform, how many animations are needed, how many joints in the rig? How long does it take to create a character and rig it up, how many do you need to have? 7 recycled units or a hundred unique characters?

Are they just low poly models with simple textures, or will there be complex materials that need to be derived from normal maps meaning there will be a high poly -> low poly workflow.

How important are other aspects of the game, like the environment, AI, particles, back end systems?  

Often the tri count is heavily dependent on testing and performance balancing as you go. What you trim back in one game to give another aspect more impact, won't be the same as another game.

If I had to take a wild stab in the dark I would start testing with quickly done stand in models that start at roughly 20k tris and see how things preformed, making sure the rest of the game had more or less equivalent stand-ins. Keeping in mind that there might be a large optimization push again later.

Woo thats a lot of questions, i didn't expect someone to ask so much

PC, desktop only, targeting gaming rigs generally

Unreal Engine. I think it has automatic LODs but i can also make them in simplygon, ill see how things go

Turn based strategy, approximately 200,

From 20-30 metres up in a helicopter view mostly, but sometimes closeups for combat. A unit pulling off a fancy killing blow is the closest thing to a cutscene that'll happen with normal soldiers, and the same models will be used for that

my plan for models here is for the combat system, where one person represents one person, and 200 of them being onscreen at once is probably a typical goal.

I'm also hoping to have an equipment system if it proves feasible, so a normal soldier will be a naked base human model, wearing an armor model, carrying a weapon model or two. I'm planning on taking all of that into account and using polycount targets as a rough budget for person+equipment combined. This may be a bit of a pipe dream, as Stardock tried something similar with Fallen Enchantress, and i believe they had a lot of performance problems with it.

Actions? mostly swinging weapons around, with the animations being based on the wielded weapon. Probably a lot of animations

Joints in the skeleton is something i'm still figuring out. My current loose plan is to have two humanoid skeletons. One being the unreal mannequin skeleton, used for a few select important people, and the second being a stripped down version of it (no fingerbones, only one footbone, etc) which will be used for the majority of normal troops. I'm not sure what else can be trimmed from that yet, thoughts on that subject are welcome

Low poly probably yes, but exactly how low is something I made this thread to figure out.

Last time i worked on a strategy game, with civ IV modding, we typically made humanoids using ~512 triangles, That was a bit ugly and low detail even for its time, so i'm looking forward to having more to work with

Materials I have little idea about, i've never really worked with them. My previous work has mostly used simple textures with gloss and sometimes glow maps, i've never yet developed for anything that has a material system. Though clearly unreal does, so its something ill be learning to get to grips with

Since its turn based, performance in most other aspects isnt really an issue, the AI will have functionally infinite time to figure out what it wants to do next and wait for player input

20k sounds like rather a lot, at least based on my limited previous experience. I know i could do plenty with that. I greatly appreciate your input though. A stab in the dark, however vague, is something I can use as a starting point.  Perhaps my answers to your extensive questioning will help refine the estimate some more

Thank you for the assistance

Alex_J

Can't thought-experiment through this. Too many variables. What the pro's are saying is you got to prototype it out to find your specific case answer.

What you are looking for is simplification, but looking for it in the wrong way. Simplify the problem not by designing some arbitrary guidelines, but rather by reducing the goal to something clear and digestible. Start with "how must the game look." Then make your models to that target. Don't take too long, just make them best you can and hold nothing back. Then optimize as much possible without detracting from the goal quality.

Then rest of the game gets built around that. If not possible to keep your quality target and 30fps (or whatever the goal is), then you got to reduce visual quality. There is simply too many case by case factors to know before-hand.

Personal anecdote:

I made characters for an action TBS, similar in design to Bioware TBS titles, and our characters ranged from 10l-20k tri's and each had 2-3 materials. Environment was just hodge-podge speed tree items mostly. Game hasn't been published yet but it runs 30fps in editor. Usually 1-2 million tri's on screen at a time. Programmer is pretty experienced, but I wasn't very experienced as an artist so things definitely could be optimized much better.

Some things we found were huge performance killers was hidden shadow casters. Like we had teeth in each characters mouth -- turn shadows off on them and it chopped out hundreds of thousands of tris. Actualyl we ended up removing teeth entirely cause no face animations, but shadows and lighting really seemed to be the biggest thing. That and materials. Definitely test out early how efficiently you can combine textures into atlases. What things can go together to save space while still keeping the resolution you need. This can only be answered with prototyping.

If you are lone wolf, keep it simple. How simple? Whatever idea you have, divide by ten. Not just quantity of assets, but every time you have some system in which one thing depends or is linked in some way to another, that multiplies workload and complexity massively. Things can get out of hand very fast when you start dipping into modular characters, user-customized shit, that sort of thing. Everybody wants that in their game, but it is a massive workfload. Not so much in the initial pass, but once you got to go back and modify things, but its all in an interweaving web, the  going gets rough.