Are 40.000 tris too much for a desktop game model?

Hello!

I would like to ask if 40.000 triangles are too much for a desktop game model or WIIU.
I think I have heard somewhere that a current AAA game (I don't remember the name) used models of this size on a console (though I don't remember the console name either... perhaps it was XBOX360?).

Thank you.

 

No it's not. I routinely use larger models with no issues on PC. Can't speak for WIIU though.

 

I don't think it is ok. Depend on what is this model. 40k for a character is way to much. but for a car in a cars game, that may be ok-ish.

 

It depends on what it is used for. Polycounts aren't nearly as much of a bottleneck as they used to be. Shader passes and texture calls are usually considered a more significant point of optimization these days. At the same time, you don't want to be throwing a whole bunch of unnecessary polygons around if you don't have to.

If those 40,000 polys are going into a unique model that isn't going to be getting cloned all over the place, I'd say go ahead. If those 40,000 polys are for a tree model that you intend to paint a couple thousand times across a landscape, you might want to think about cutting down those numbers.

For character models these days, 40k is not considered unreasonable on desktop systems. (possibly even some mobiles, again, on a case-by-case basis) Those extra polygons can be used to smooth out areas that might otherwise look blocky, and assist with deformations around joints.

 

What's the model of and for what kind of game?

 

40K polys is nothing for a modern day PC, times that by 10 and it's still nothing. As Richard said, if you had 1000X 40K poly tree's amongst the rest of your scene then you're pushing your luck, I think the terrain alone for the demo I did was 1.5 Million polys and that ran 60+FPS no issues.

Try that on an old mobile, it would sink it but on a modern console it's fine. So it depends on what you're doing..

 

I forgot to mention that I am talking about a game character, not a tree.

 

Well you should be fine then (for PC anyway), not sure about the WII U. I don't deal with it..

 

Without giving any desktop specifications I'm going to assume it is a 10 year old desktop just for the sake of keeping it on par with the Wii U which is basically a GameCube after drinking some espresso.

In which case, yes, 40k polys for a character is too much and completely out scope for the project.

 

In the end I think it kinda comes down to how well optimized the game is in the end, and what kind of machine you plan on using to run the game. I think if you were to optimize models with distance (Whats actually noticeable) with view distance scripts or what is refered to as LOD (level of detail), then you would find that 40k polys up close would not be too much for any modern PC. As with anything though, it truely depends on the scenario in which your 40k poly model is being used, and what kind of load your engine is already under to determine the resulting performance. (Scripts, amount of models, lighting, shaders, other things that could impact game performance).

WII U, No clue.

 

https://forum.beyond3d.com/threads/...st-blog-entry-with-interesting-numbers.39321/

 

Be careful with the tris/vertex amount. I have several models that show up as "13k" in the fbx but inside the game-window it has WAY more.

Also, if it's only the localPlayer this is fine, but when you have the same model for remoteplayers this will fill up very quickly.

My Mesh shows in the preview window with 10k verts, 11k tris, 4 submeshes, but in an empty scene with only the model it got 24k tris + 70k verts ... not sure what's causing that.

 

UVs will often count as more in the total.... each seam will be 2x the number of geometry vertices...

so, you try to make the fewest # of seams as possible... which can be difficult for certain things

 

Seems like 40k might be too much for the average humanoid - not because it's a strain on the system, but because it could be optimised a few thousand polygons and still look as good. I think that should be the goal: Optimise it down until just before it starts looking crap

 

So is it just one character, or will you have many such characters?

The poly count of one model with no context isn't of much use. You need to consider the poly count of the scene, and also what you're doing with the polys.

 

The lesson here, is that context is king. Adjust your expectations for polycounts based on the different factors you're dealing with. (complexity of the scene, most common camera angles, how many characters at a time, complexity of animations, etc...)

I come from the old-school. I started learning to model back when the common objective was to shave down every model to the bare minimum number of polygons possible. Back then, it was considered necessary to get every game character under 800 polygons each. Any more than that was pushing it. We would gut our models over and over, squeezing them for every iota of detail that we could get out of that minuscule geometry.

There are a lot of great techniques and principles that you learn from an exercise like that. The importance of silhouettes, optimization for animation, the ability to maximize detail through painting textures. I saw some fantastic art come out of those incredibly strict limitations.

These days, you don't have to be nearly so conservative, especially when it comes to polygons. Most modern 3D processors are optimized specifically to handle an unholy number of polygons. Millions of polygons per scene aren't considered that big a deal any more. For the average game character, 40k is fine, even on the Wii U. Just make sure you aren't throwing a lot of unnecessary polygons around. Give it one more geometry pass to cull a few extra faces that aren't really essential. No need to gut your model anymore, but why tempt fate when a quick pass can cut back a few thousand polys?

 

Well, I attempted to do some reading on it and while I couldn't find some more concrete numbers I read some people mention that the Wii U cranks out about 250M polygons a second and the PS4 hits around 1.3B.

Those numbers could be WILDLY off since I could only find them in other forums. So if anyone could correct me with better sources please do.

Anyway, you also have to remember that your 40k poly model has to be drawn at least 30 times a second. so 30 x 40,000 = 1,200,000 polygons per second. Divide that into 250M and you get 208 copies of that model at 30 frames per second in ideal conditions on the Wii U.

That said, you're not going to have ideal conditions. The temp and humidity won't be ideal. The system will have wear and tear on it. Whatever.

Also, you have to remember that the vast majority of PC gamers have rigs as much as 10 years old (despite the loud minority). I think better to err on the side of caution and aim for around 150M polygons if you're looking at a low budget title. But that's just me.

Now if you're looking to do a gussy'd up graphics monster of a game, then yeah, shoot for a billion.

At any rate it really boils down to what that character is doing. A game like Street Fighter 4 you could easily do 40k. While a game like Total War, not so much.

Context really is everything.

And like Orb said, optimize.

 

Build a prototype and simply try out. There is no other way to find out if the poly count is too high for your current project. Because poly count is ways not the only factor. It's in fact one of the smallest factors. Modern desktop graphics cards deals with the pure mesh data in an eyeblink, even in megapoly range. What eats most computing time is lighting, shaders, game code and physics.

So simply try out. Every game is unique here.

This assumes though that your modeling skills are good enough to create just the needed geometry. Means that your mesh is not blown up by not needed loops and faces, and has a clean topology.

An image to judge could help here

 

Depend on the how many polygon you plan to use for the rest and which shader you use.
40K polygons is not much even for the lousiest cheap mobile phone now, but if the shader has dynamic lights these polygons are rendered per each light.
Like it was said, it's not that it's too much but that it's not needed. Even at 20K you have a very detailed model, and you can use these saved polygons elsewhere.

 

Thanks to all of you!

 

Hope this helps.
Some of these are from Beyond 3d forums and other parts of the net.
Special thanks to Cops and Rappers and others from beyond 3d forums.

https://forum.beyond3d.com/threads/...st-blog-entry-with-interesting-numbers.39321/

The DOA models are the highest polys I've seen in a fighting game so far.
Mila's and Kasumi's polycounts are ridiculous!!

Vanessa Shneider PN03 Gamecube = 14993 tris.

Rachel DOA 5 = 46152 tris

Vanessa Lewis....... (now a fair skin Black girl ) from Virtua Fighter 5 = 14429 Tris

 

Trying to find the polycount for the Bayonetta 2 characters.
For the Wii U.
I know Platinum games mentioned that Bayonetta herself, from Bayonetta 2 on Wii U, her polycount is higher, than the Bayonetta model from the first Bayonatta game.

 

Thanks for those visual examples, BrandyStarbrite! It helps to see the triangles themselves, it can give you a better idea about the underlying structure of the models.

Note that the more optimized models tend to concentrate geometric detail in specific areas, and simplify the number of polygons in areas that are flatter. This is a matter of optimization. For instance, all of those models have a considerably higher concentration of polygons on the faces. All of them are designed for some degree of facial animation, and need more polygons in those areas in order to compensate for animation, and zooming the camera in close to the face.

 

Yep.
What you said is definitely true dude!
You hit it on spot.
Right on target.
Looking at the models in the pics does make you realise that.
You also get an idea, of how Japanese game developers, model their characters for their games.



PS: I also, forgot to mention, that many game modellers, including that UBISOFT guy,
who worked on Assassins creed game said.
Half of the polys are for the face, the other half is for the body!

 

While true, I imagine it is also occasionally frustrating, especially for those Assassin's Creed guys. Facial polys are important. The face is one of the most expressive parts of any character, so having extra polys in that area makes sense. At the same time, a lot of games actually take emphasis away from the face during play. A third-person game like Assassin's Creed normally has the camera behind the character, where the face is largely invisible. Also, in Assassin's Creed games it's common for the character's face to be regularly obscured with a hood.

I imagine that might be a little frustrating for the artists. They go to so much trouble to make their character's face expressive, and then it spends most of the game in shadow, out of the player's view.

 

Excluding the Wii U for a second, in todays setups there is little reason to struggle or obscure unless you're aiming for last gen / 10 year old hardware or mobile. Especially if you have an RPG type game where face FX / Deformation and blend shapes are important.

Don't get me wrong it's no excuse for sloppy extensively over polyed artwork, smoothing groups etc. is still a good idea. But today the budget has largely increased, I think one game on console was recorded at having 11Million polys in view at any one time at full frame rate.

The main issues I have relate to tech, you really still need to lightmap and correct material complexity.

 

Yes, a substantial amount of this discussion is honestly a bit moot. While general optimization is a good policy to follow, modern hardware and its approach to polygons make most poly-limits a bit extraneous. Even twice the quoted amount (80k for one character) wouldn't be unreasonable. The triangles simply aren't as important to optimization as they once were. Tweaking the shaders, textures, and object pooling is far more likely to result in improvements in performance. Shaving off a few extra thousand polys won't make the difference that it once did.

You kids these days don't know how good you have it.

 

Well I remember having an 80286, which had a processing speed of 6MHZ! Also had a ZX speccy, so I'm not that young ..

I think Ryse's poly count for Marius was 85K on XBone and 150K on PC. A lot of the min specs for modern PC games are getting ramped right up, it's not unusual to see min specs around a GTX 660 and they have quite a thump in terms of power.

 

Generally speaking, other stuff like draw calls and whatnot will hurt you far more than raw polycount. We have so much processing power nowadays it's not even funny. Put it this way, the dragons in Lair were about 150k and the cars in Gran Turismo 5 were about 200k...and that was on the PS3 from 2006. Some characters from Virtua Fighter 5 (2006) were about 40k.

http://www.gameartisans.org/forums/threads/23520-Historical-Poly-Counts

It mostly comes down to the type of object you're making, generally speaking you can go higher with important "hero" objects, while less important environment objects, or objects that will appear in large numbers, should be lower. To use sports games like FIFA as an example, the actual soccer player models might be relatively high poly...but all the characters in the crowd would be extremely low poly-heck in the old days they'd be billboard sprites.

That's because the devs know that there will always only be X number of hero characters and X number of crowd characters visible at a time, and so they can budget accordingly.

 

How does this LOD for game character models work? I have read that quite a few times now.
I imagine it the following way: When a model is still far away, you load its very low-res model (LOD2). When the model gets closer to the player, you load LOD1 and dispose LOD2. Now when the model gets upfront to the player, you load LOD0 and dispose LO1. But loading a LOD0 model takes enough time to cause a visible delay, I guess.
Did I misunderstand that approach? Are all LODs loaded upfront, and we only switch between them?

 

That's typically how it's done, yes. Loading and deleting during runtime is not very efficient.

 

I think it's worth a quick reminder that LODing is a thing. Just because a base character is 40k polys and 20k are in the head doesn't mean that it has to be that way all the time!

If LODing isn't generally needed for your game then you could have switchable close-up and distance heads. Use the close-up for conversations or other facial animation, and the other for running around etc. Save most of the polys most of the time, and only show the extras when they're in use.

Alternatively, your 40k model could be LOD0 and have a 20k face, followed by a 22k LOD1 where the only difference is a lower detail face, followed by LOD2, 3 and 4 with lower details for distance/crowd rendering/etc.

In that context, even on a Wii U it might be perfectly alright to have a couple of 40k characters in dialog scenes where the detail is useful.

 

Ahhhh,.........found a Bayonetta polycount reference pic.

 

Oh my!
Sorry for forgetting, to answer the Wii U part of the question.
My bad!

The character models (or a single character) in TTT2, on the Wii U, PS3 and Xbox 360 are close to 20,000 tris.
Jun Kazama is about 19262 tris.

PS: I could post a pic of her (Jun) I found, if you want @TobiUII.

 

Heh! So true.
That's the same thing I thought.
It's kinda like, all of that hard face modelling/facial expression animation work, for almost nothing! He! He!

 

As previously mentioned, it depends on the type of game as to whether 40k is acceptable. Also depends on the games optimizations as well.

For example, if you were creating an MMO like World of Warcraft they use character models that are 5k - 9k.

 

And you didn't mention.....
What type of game are you planning to make TobiUll?

 

He is Groot.

Anyway, I think it isn't too much, certain AAA titles such as GTAV have character models that can be up to 100k polys. So not a biggie I think.

The thing is though that you should always make as simple model as possible - if you can shave off several polygons without noticeable loss in quality, DO IT.

 

A 40k model should have some very impressive details on it.