in search of reasonable triangle limit...

I'm starting to take for granted my 2Gig of RAM, P4 processor and Radeon X1600 Pro card and happily render a million or two triangles without too much of an impact on frame rate.

But in the real world what's a reasonable limit to put on triangle counts in a game these days?

Should I just not care and assume that as computers naturally get better that if it runs on mine it will run on other people's, if not now then at least eventually?

In the real world. Humm, I think I could draw 1 maybe 1 1/2 in a second. hahahahahahahahahahahaha.

No useful comments, but am interested as well. I always assume that mine will be below adverage by the time I finish.

LOL! that's a good one :O)

That's like how long is a piece of string? You can't possibly generalize on how many triangles a game should consider reasonable.

Use as many as you need and no more. It sounds trite but that's what you have to do. You have to make a judgement about how close things will be to the camera, at what point does adding polys stop making it look better? At what point does reducing polys start making it look worse?

I remember from earlier (of course depending on the engine), from 40-60k triangles on an average rig..

note that I'm not 100% sure on this, it's just something I seem to recall..

My new system had tons and tons of power, shame I will not be able to see what B3D could have done with that 8800 GTX. One thing I wanted to test was the amount of geometry B3D would happily shift with it.

Well for the things I'm doing, adding polys will pretty much always make things look better. Also the more polys I can display, the less LOD work I need to do. . . and hence less effort it will be to code. That's why I need to work out how to balance things.

I need to make these descisions based on a hypothetical "average" PC, not a top end gaming pc.

Yes, there's more at play here than just the number of triangles because a lot depends on what other code the game is running.

However, what I am asking people is this:

Roughly how many triangles do your games/projects tend to end up with after code/LOD optimisation is put in. I know this will vary a great deal but the rough average number is probably a good guideline to use when it comes to deciding how much LOD/optimisation work I need to do. . .

I'd be more worried with what exactly you plan to do with said triangles.....and how many surfaces are taken into the question.

I find that it's pretty much what Gab said, use whatever amount you deem reasonable to the project without being overkill, or underkill in terms of overall visual look.

A good balance between surface count and number of triangles in a surface is probably what you want to aim for. You'll need to do some test's however, determining what's right for your game.

I'm sure I read somewhere that B3D uses lists for triangles. In that case its not going to be as fast as something that uses VertexArrays.

Anyway, LOD is your best friend. Same with Bump-Mapping. Bump-Mapping can give close up objects a nice illusion of fine detail, and LOD can remove triangles that would otherwise not be noticed in objects in the distance. I'd say around 2,000 - 3,000 polygons per character, 10,000 - 20,000 for the level etc. Your probably looking at a complete scene being around 40,000 - 50,000 polys. It also depends largly on what your trying to create.

For RPG games you hardly ever get relatively close to an object so some detail can be left out. In FPS games your often right up against objects/characters/scenery so fine detail is essential to make it look good.

Now don't quote me but is not the number of different textures for each surface composed of triangles ALSO a speed factor delimiter???

Simply if you had a cube made up of say 400 triangles (why I have no idea!) and only 1 texture to render that spinning cube... would it not be faster than if you had 20 textures for each 80 triangles... or 100 textures for each 4 triangles etc???

I really doubt that. Maybe yours is the first, but I've never come across a game yet where it was.


I respectfully disagree. I think this will send you right up the wrong path and you'll either have a game that could have looked tons better or you'll have a gorgeous game that runs like a slug in treacle. You're not making an average game, you're making YOUR game. The only worthwhile way of finding out how many poly's you can push is to try it.

Anyway, you're clearly intent on working from an arbitrary average and I've said all I can to caution you against making a mistake, so I'll leave you to it. I genuinely hope you prove my concerns to have been misplaced.

As a rule of thumb, I don't like my level mesh to go above 10,000 polys - but it's fairly simple geometry and it's pretty much visible all in one render (no occlusion) with a distant camerarange involved.

Of course surface count is a killer, but my feeling is that Blitz's 3d rendering is not overly quick (compared to other engines) - so it's better to underestimate/optimise I guess.

Okay, I think I may have been too vague and general here. . . perhaps I should give some background to my particular thought process. . .

Well, my current project is an experiment to display a massive life size detailed terrain at large distances.

Given that bumpmapping is not something I've done before and don't really have time to learn, using large numbers of triangles is really the only way I can think of to improve the amount of detail rendered. Currently the LOD system (with 3-4 levels) is giving good visual results with around 100-150k triangles but I fear this may be too much for many cards especially if I also include other graphics such as trees etc.

Anything below 100k triangles starts to give unattractive results, so I need to know what upper limit I should be setting. This will go on to determine the amount of detail I can show and also the maximum camera range.

I know I should really be giving the option for someone viewing the terrain to set the amount of detail according to their card capabilities and I probably will. But I will need to figure what range of settings to offer.



Probably right there. I guess I am just looking for an easy way around the fact that I have only one PC to develop and test on. I spose I will only find out how well my code works on other machines when I release something. . . if I ever get around to it! :O)

I say about 50,000 - that's what I get a good frame rate on my Radeon 9600 XT with antianalizing cranked up all the way (in the Catylst Control Center) , and I can get a good 50 fps with that, but with antianil off, i can get 800 fps (on the dragon demo)! lol...but if you are worried about a poly limit, you can use EntityAlpha mesh,0 or VertexAlpha# (curface, index) ...

Are there monitors that can cope with this???

> > 800 fps
>
> Are there monitors that can cope with this???

Are there eyeballs in eye sockets that can cope with this???

..your concern should be number of surfaces and textures assigned to them...on my old nVidia 5600FX I pushing without any problems 200K polys with 50-60FPS...but once textures start appear over geometry, things is different...so, number of textures assigned to your fancy polys will determine real data transfer and speed you need to achieve in order to feed properly your monitor with nice, soft, decent FPS...

Unless you're trying to recreate that age old classic "Attack of the sheer number of demon killer triangles on screen", I think you should worry about the game instead.

True, number of surfaces is very important too. Also, TrisRendered doesn't have to be TrisInWorld.

Basicly I'd say use a max of about 40 k Tris. This will be playable even on cheap hardware.

But it really depends on your target customer class. Do you want to challenge Oblivion or Doom3? Then you surely may use the latest and best hardware. If you want to have a max of potential players then you should use lower spec requirements.

is there a limit with DX7 ? or is that just a myth?

i think the real impact on fps is how many animated objects you have on screen.

Pete