Model Polygon Counts

Level design, models, animations, physics, etc.
  • 1
  • 2
11 posts Page 1 of 2
XIXWYRMEXIX
Posts: 54
Joined: Sat Mar 07, 2015 3:13 pm
by XIXWYRMEXIX » Sun Jun 02, 2019 4:47 pm
Hey, I was wondering what polygon count I should be looking towards when finishing a model to be "game ready" in regards to torque. I realize this count changes depending on hardware, hardware advancement, game engine advancement, game type, etc. But as a basic rule of thumb for torque 4.0, in an FPS style of game what numbers should I be looking for? I do not want to for example, be making a 10k polygon model when I could be (or should be) using 20k, or 5k etc. etc. I simply have no idea what I should be looking towards building, I am new to making models. I also want to scale all of my models accordingly, for example I own models that are not made by me, I would rather upscale them or downscale them depending on what I make my own models. Any pointers would be appreciated, thanks!
Duion
Posts: 1357
Joined: Sun Feb 08, 2015 1:51 am
 
by Duion » Sun Jun 02, 2019 8:51 pm
Well first it depends on the game, second with modern hardware it does not matter that much if your game is not that large scale, you can get away with lots of very poor optimization nowadays.
To make it short: 10k is good for an FPS model, 15k is high quality, above is high end, 5k is kind of low end now probably and below it is more for stylistic intentional minimal stuff.
To make it shorter I aim for 10k. The Torque soldier has 15k, but he also has lots of stuff on him like armor plats, holsters, grenades, pistol etc which adds a lot, so "naked" he would be around 10k.
But a good artist often makes always high quality and reduces later depending on the need, upscaling on the other hand is not recommended in most cases, since it does not add much "real" additional detail.
Jason Campbell
Posts: 331
Joined: Fri Feb 13, 2015 2:51 am
 
by Jason Campbell » Mon Jun 03, 2019 5:53 am
Keep your eye on drawcalls.

Use metrics in the console to see.
metrics( "fps gfx" );

Try to keep models with single textures. I think adding normals and specular adds a draw call for each. Shadows eat performance a lot, even on decent hardware.

I think 1000 is high but others have said 2000 is good.

Read this post Steve, covers quite a bit
http://www.garagegames.com/community/fo ... ead/112477
Duion
Posts: 1357
Joined: Sun Feb 08, 2015 1:51 am
 
by Duion » Mon Jun 03, 2019 9:56 am
Yes drawcalls is more important, think of it as in painting, polygons is like complexity of the painting, but a drawcall is where you need to change the color of the brush.

I heard of the myth, that 400 polygons are "free" due to the cost of one drawcall, so each model can have at least 400 polygons theoretically before it costs more. However my tests showed that less is always better, it is just not that relevant probably.

What I do when developing is I only start to worry about optimizing my scene, when my fps count gets below 60 fps on max settings, I have a mid range gaming PC and this is what I target for, so that anyone with a halfway up to date PC can play it flawlessly on max settings, I also have options for really low settings that can make it run on almost any PC.

A typical scene has around 3000 drawcalls and maybe 3 million polygons including shadows.
XIXWYRMEXIX
Posts: 54
Joined: Sat Mar 07, 2015 3:13 pm
by XIXWYRMEXIX » Mon Jun 03, 2019 4:06 pm
Well first it depends on the game, second with modern hardware it does not matter that much if your game is not that large scale, you can get away with lots of very poor optimization nowadays.
To make it short: 10k is good for an FPS model, 15k is high quality, above is high end, 5k is kind of low end now probably and below it is more for stylistic intentional minimal stuff.
To make it shorter I aim for 10k. The Torque soldier has 15k, but he also has lots of stuff on him like armor plats, holsters, grenades, pistol etc which adds a lot, so "naked" he would be around 10k.
But a good artist often makes always high quality and reduces later depending on the need, upscaling on the other hand is not recommended in most cases, since it does not add much "real" additional detail.
I know it depends on the game, I said as much in the OP, it was a "basic" rule of thumb I was/am looking for, your short and shorter answer is exactly the info I wanted. Thank you I appreciate it Duion. I do know a bit about making models I have just never done it myself so I have "book" knowledge about the process but no experience. For example I know you make the high quality then go down to the game quality, this is a lot like pictures and graphics for webdev which I have done a bit of. Rarely does scaling up go well as far as quality goes. However I know it is possible, by adding more details to the model itself, then scaling back down. Not saying I am doing that or want to, just that if I needed to I could. My models I own are in the area of 10-15k though so I do not think I need to do anything like that.
Keep your eye on drawcalls.

Use metrics in the console to see.
metrics( "fps gfx" );

Try to keep models with single textures. I think adding normals and specular adds a draw call for each. Shadows eat performance a lot, even on decent hardware.

I think 1000 is high but others have said 2000 is good.

Read this post Steve, covers quite a bit
http://www.garagegames.com/community/fo ... ead/112477
Thanks for the info Jason and the link to Steves post I will check out and read. I love Steves tutorials and info, always full of good stuff and usually interesting and fun to read. I have done a lot of his older tutorials. 1000-2000 seems low to me (from my research anyway), can you clarify? I will of course be checking the FPS counter and the console metrics, those will help a lot!
Yes drawcalls is more important, think of it as in painting, polygons is like complexity of the painting, but a drawcall is where you need to change the color of the brush.

I heard of the myth, that 400 polygons are "free" due to the cost of one drawcall, so each model can have at least 400 polygons theoretically before it costs more. However my tests showed that less is always better, it is just not that relevant probably.

What I do when developing is I only start to worry about optimizing my scene, when my fps count gets below 60 fps on max settings, I have a mid range gaming PC and this is what I target for, so that anyone with a halfway up to date PC can play it flawlessly on max settings, I also have options for really low settings that can make it run on almost any PC.

A typical scene has around 3000 drawcalls and maybe 3 million polygons including shadows.
This is good to know as well. I will attempt to keep the draw calls in check. I will also attempt to use your rule of thumb as it seems reasonable, I also have a mid grade machine right now, I will worry about optimizing for lower end machines after I have gotten the process down.

Thank you both for the responses, I appreciate it a lot. I will post back after I have played around a bit with your information and advice.
Azaezel
Posts: 461
Joined: Tue Feb 03, 2015 9:50 pm
 
by Azaezel » Mon Jun 03, 2019 7:51 pm
Since you haven't specified player, weapon, vehicle, or background asset:

For physics, vertex density means more than sheer vertex count on the asset side. Particularly if you're skipping making collisionmesh proxies and usings straight what you see is what you get collisions.

Per scene we've found 4 mil or more ends up a high 30s given our end has a high amount of object-instance overhead (and working on eyeballing ways to tone that end down) which seems to be the biggest killer of perf (again, this end).

Call it 20-30k highest LoD meshes for enemy units (up to say 64 on screen at a given time) with fairly agressive LoDs (5 or more stepdowns)

Player side we're using 40-60ks for player+gun (fewer players than bots) for 3rd person, with 10-20kish say for first person weapon presentation.

That help?
Duion
Posts: 1357
Joined: Sun Feb 08, 2015 1:51 am
 
by Duion » Mon Jun 03, 2019 8:11 pm
Yes vertex count is more relevant, polygons are just the traditional measurement, you could go really into nitpicking mode and argue how slanting edges are harder to compute than straight ones, but it becomes very relevant in collision. However I use the visible mesh for most static scene shapes, because there is hardly anything you can LOD there on the collisionmesh, a low detail collisionmesh is something for organic objects like player characters, like the default player in Torque is just a box.

More helpful maybe would be a showcase thread where you post your models or scenes and others can say what could be optimized. To argue blindly without knowing what it is about is a bit complicated, but I can identify crappy art pretty much instantly, when I see it.
XIXWYRMEXIX
Posts: 54
Joined: Sat Mar 07, 2015 3:13 pm
by XIXWYRMEXIX » Tue Jun 04, 2019 2:24 pm
That helps a lot Azaezel! This gives me an idea on what you can do based on a lot of metrics in regards to your counts and setup. Out of curiosity what types of high amounts of instance objects are you dealing with? Scenery? Foliage? Buildings? Which leads me to another question, if you have say 10k count player models, with roughly 2k count for weapons, at that level of count what would you look at for trees and buildings? Obviously a rough estimate is what I am looking for again. Also what hardware are you running?

I haven't specified any player, weapon, or vehicle asset because I haven't decided on which/what I am using or going to use. That is the whole point of this thread, I am doing research to determine what rough constraints I am looking at so I do not have to go back and re pipeline everything because I made a mistake on model designs. Or at least minimize that. I want to have a plan BEFORE I go hip deep into a model frenzy.

@
User avatar
Duion
See above. I do not see any one arguing. Different games will of course have different hardware requirements and different model densities. This is just a discussion on rough capabilities that different people are using so I can determine which direction I want to go with the models I have, the models I will generate, and the models I will personally create. I like having a plan before I do anything, and I like to research before I form a plan. I will make a thread when I have determined which direction I want to go and actually generate more assets.

Thus far I am looking at Duions model density. Roughly 10-15k player models, with roughly 2-3k for weapons. I am not looking for some triple A title type game, the plan is a simple sci-fi shooter, with mechs, possibly some aliens. I am looking at an open world as opposed to a level based game. The "mechs" are going to be smaller than "traditional" mechs, a step up from power armor size wise but not like gundam, robotech or battlteech sized. Think more heavy gear/votoms. I already have those designed and drawn up, I just need to make the models for them. I am at the stage to begin doing this, thus the thread. I also have a full storyline for the world and the game all written up. So I am not floating in a vacuum here, just asking random questions with no real idea of what I want. There is a story line, a "universe" history and back ground, drawn concept art, some game assets already owned, universe characters created with written story arcs.
Duion
Posts: 1357
Joined: Sun Feb 08, 2015 1:51 am
 
by Duion » Tue Jun 04, 2019 4:46 pm
If your "players" are mechs, you can just make them up to the maximum limit of ~65 000 vertices per model, mechs are bigger and more complex and make up a lot of your game visuals, so go as good as you can, for the distance you still have LOD levels anyway. At least I would make it that way and to optimize performance I would rather reduce the world around and remove some trees or so.

Modern games sometimes use that high polycount models for closeup or cut-scenes, it depends how many of them will be in the scene at the same time and how much else will be going on in the game at that time.
XIXWYRMEXIX
Posts: 54
Joined: Sat Mar 07, 2015 3:13 pm
by XIXWYRMEXIX » Wed Jun 05, 2019 3:53 pm
The thing is Duion not all of the players will be "mechs", it will be a combination of standard sized players "soldier models" (which I will use as stand ins until I make my own) and "mechs" (which will not be like traditional mechs, I am making heavy gear/votoms sized mechs. These mechs do not stand as tall or big as mechs from other genres. They are roughly only twice the size of a game world person. So roughly twice the size of the torque soldier. so if the torque soldier is 6 feet tall my mechs will be 12-15 feet tall. Do you think 30k would be about right for this reduced size and in comparison to the standard size player being 10-15k? Also what do you suggest for tree and building density for models? I would love to hear your opinion on that.

Example of size-

Attachments

armored-trooper-votoms-anime.jpg
armored-trooper-votoms-anime.jpg (111.27 KiB) Viewed 1498 times
5741e48781cfaaaeefd5873107e4960d--super-robot-mechanical-design.jpg
5741e48781cfaaaeefd5873107e4960d--super-robot-mechanical-design.jpg (67.35 KiB) Viewed 1498 times
  • 1
  • 2
11 posts Page 1 of 2

Who is online

Users browsing this forum: No registered users and 2 guests