Torque limitations

#1 A) Depends on the game you're making. First person shooters will vary the most in the poly range, depending mostly on how many guys you want on the screen at once. RTS' and the like have generally lower polycounts so you can cram hundreds or thousands of units on the screen all at once. Fighting games that have 2 or so people on the screen can have very high polycounts, because you dont need many people.

B) By default, Torque is pretty capable at rendering stuff, but if you get too much too fast, it buckles(though this goes for all game engines). Now, if you're going for a run of the mill FPS, with stock TGE code, it's a best guess to go around 4,000 for a player, and a few hundred for weapons.

C) If you go about modifying the rendering code to do batching, vertex buffers, or whatever else optimizations you'd add in, it can drastically boost your rendering power letting you do more with less.

D)Dealing with the computer:
If you intend to release/sell the game, you need to pick a base level of computer power to base your minimum requirements off of, and base the rest of the game's asthetics off of that. Pick a general minimum requirement, and work off of that, Dont use a highly specific computer configuration either, that doesnt allow for specific quirks of different cards or processors to be handled as well.

Summary: If you're doing a run of the mill FPS-level polycounts, with no real modifications to the rendering code, somewhere around 4k for players, and a few hundred for weapons.

#2 Thank you Jeff, Now how may polygon can torque have on a screen at once. Because i wan't to try to make a game on the Xbox live arcade, so can i get it up to 100k on screen or from what i heard only 60k?

Thanks in advance again.

#3 This again falls back on what all you're intending on doing.
In a top down camera setup like an RPG or a RTS, you dont have as muhc draw distance to worry about, and can use that to cram more detail into the scene.
For FPS' and the like, where you have a sizable draw distance to worry about, you'd need to take other things into consideration.
First, distance fog. This will help considerably if used right, because you dnt have to draw stuff 5 miles out.
Next is LOD's.
Torque allows you to switch between LODs on models, which means that you dont always have the high poly models to worry about, so with plenty of LODs on things, you can have more stuff in-scene, while still being playable.
Total polycounts is a lot harder to estimate for what Torque allows for, because it very heavily banks on what the camera's doing, as well as how your LODs are and how optimized stuff is made.
If you stay smart when designing your models and the like, i can see Torque pushing 100k without too many problems, though if you need several hundred thousand polys, you need to start looking into optimizing the culling of the world and special optimizations for models and the like.

#4 I haven't seen any hard polygon limits in Torque yet. And I doubt there are any in the engine. In my testing, I've had around 50,000 - 60,000 polygons on screen at once with decent frame rates (in the mid 50s). One of these days, I might try it with a higher total polygon count.

On a related note, with the current generation of DTS exporters, there is a 10k polygon max (IIRC) per mesh. You can get around that by breaking up large objects into multiple sub-10k meshes. The next batch of exporters will no longer have that limitation, from what I hear.

[Update]

I got curious so I gave it a try. In my quick and unscientific tests, Torque continues to work fine at 102,000 polygons and above.

#5 It depends on what you want. Do you want multiplayer access? Do you want deformable geometry? Do you want physics? Do you want to ghost those elements out to clients?

It is so much more than simply rendering a high-poly model with LOD. If you don't want to do anything with the model, TGE can have high-poly models (exporter or coder depending). If you want it to do anything, you will have to make different compromises. Just like any engine from Wolf3D to Unreal 3.