Will the engine hold up?

#1 Do a forum search for Speedtree. and also look in the snapshot gallery back a few pages. Right now Iam too lazy to post a link.
But I know you will be please to see the result. :)

John H

#2 I have seen that, but I mean With out buying 3rd party technology. I mean I have seen the SpeedTree RT demo from their site and quite franly it runs hirribly slow. the trees look good but you get about 7FPS. What I'm asking is that if actual objects and grapsplanes were used and place in a mission at the level that they are in the new battlefield engine is torque gonna be able to hold up against it or is it going to crawl?

#3 If I'm not mistaken SpeedTree RT is a couple grand anyway

#4 Horribly slow? What kind of system are you running... Which demo was it? The Nvidia one I noticed is kinda sketchy (due to some non speedtree related code I believe) for some reason, but the one called "The Valley" and the other massive forest ones all were blazingly fast for me.

#5 Not sure it was a while back. I tried them on a Nvidia FX 5600 256mb with a P4 2.4GHz 800FSB with 1GB Dual CH DDR400. But quite frankly I can't affoard their pricing. I can hire several people to place the objects by hand for the price of Speedtree RT.
$6k for trees is a bit rediculas. I will goto their site though and give that demo a test run again and see if it still runs like crap.

#6 Melv has a some nice code that handles objects differently from the normal torque route. But since he got a very very important special event coming up in his life I couldnt say when it will be due. :))

I misunderstood your post James your asking about the performance of torque with a lot of objects. My bad :)

Look for the stratgem (forgive my spelling Melv heh) snapshot in the gallery and you will see where he talks about it.

John H.

#7 James Brad Barnette

I think Torque can do what you are asking for and a bit more.
Just not out of the package. You'd have to add some code snippits, it'll look great - trust me. :)

I'm the level creator for our project and we use all the replicator code, with some cheap tricks and alot of sprites. It looks IMO better than BattleField and the performance is better.

#8 James,

Since you have the engine you can easily test this. Just go into the mission editor, add a fxShapeReplicator object and set the number of objects to however many you want.

#9 Stefan: looks better than BF:1942 or BF:Vietnam? becasue I must say BF Vietname looks pretty good. Do you think you would post or send a screen shot?

#10 Man Tried the Speedtree Demos again and they are still dog slow. I tried the on a quadro FX 3000 which is the highest end card they make and it still crawls.
Tried the Vally and the Nvidia demo. the Nvidia one is crap it freaks out with branches tured on. I swear it seems that Nvidia has gotten really bad about breaking sstuff constantly and then patching it a couple weeks later.

Sometimes I think I should have gotten the Wildcat instead.

#11 They run great on my dev system at work (which has a consumer level nvidia card in it). The Valley runs 30hz at 1600x1200... Very smooth. I think they have it optimized to run at 30fps, as that was the same framerate I got at 800x600. :)

Trying their other demo now.

#12 Other demo is a bit messed. Rebooting and trying it again.

After some fiddling, it looks like some of the shaders related to wind/fog are messed. Turning those off makes things work well. Runs smooth and pretty.

#13 James

I don't like SpeedTree either, it seems very CPU/GPU intensive.
If you check the screenshots section out and browse some resources, you'll get a fairly well picture on how stuff looks.

I already posted a thread about many nice pictures in the forums, go look for it and you'll find it. It's very nice pictures :)

#14 Yeah Well I don't need the wind an all that stuff anyway. I just wanted to know if I made trees and plats with mutiple levels of LOD would I be able distribute them as thickly as Battlefield Vietnam and maintain at least 1024x768x32bit at least 35-40fps with say a GF4 ti4400 128mb or better. Becasue by the time I get the game out the door that will be a very old card.

#15 Hmmm.. I don't know actually.

I used around 36k billboards in a pretty tight area, all 128x128 greyscaled, some trees with LOD meshes and the normal fog, two-three players. Yeah, that stuff.

I got around 15-20 fps all the time, which sucks :P Yeah I know.
This was with 1024x768x32bit using GF4 ti4800 64mb.

Why don't you just try it mate? It's so much easier for you to decide upon this yourself than to let others try and talk you into it :)

I believe what you're asking for can be done.
But you will get a better and more real result if you try it yourself.

Good luck.

#16 After reading this post i tried creating a forest using fxShapeReplicator with the tree model in starter.fps. i put 7000 trees in a 600 radius circle. as many of you can imagine fps drop to the equvilent of Myst!! but i found that if i reduced the visibility to 100 and the fog distance to 10 the framerate becomes much better. not perfect but passable. around 19-35 fps in the the most crowded parts(I have a Radeon 8500 128MB for those that want to know)It looks pretty cool actually. the effect is not perfect though. in places where the trees are sparse you can see the trees and landscape popping into existance. and outside the forest the short visible distance becomes painfully obvious. but inside the dense parts you can't see this.

If the rendering of dts where a little faster this would be a very good solution.

i had a problem with collision though. The Shape replicator seems to be displacing some of the models collision meshes. I found myself walking through trees and bumping into trees that weren't there.

#17 The rendering of foliage in realtime is one of the most difficult things to achieve accurately, second only to water rendering. As GPUs get faster (and boy they are), things will become much easier as options open up but at the moment, when trying to target a broad range of hardware, you'll need to develop algorithms to produce such effects.

It's unreasonable to expect to have an 'object' that you can throw into a mission and say "give me 10,000 trees" and get 'optimal' performance.

There are so many techniques available and each has their own pros and cons according to what you want to achieve. If you're just an artist and you don't understand the technicalities of adding foliage then you'll need a programmer to advise you.

You could use billboards (such as the fxFoliageReplicator) and live with the fact that up close, you've only got 2D billboards as well as all the options of how the billboards are aligned e.g. camera or world. You could have intersecting billboards that are world-aligned that look better from various angles and have a tendency to fill-out the scene easier. You could use 'imposters' which are derived directly from real 3D shapes which give more realism as well as potentially using depth-imposters to allow for planar depth comparisons in 3D space. You could use real shapes and provide many LOD for better transitions but you'd better be sure that your shape rendering class batches the data efficiently and doesn't include too much state chuff to get in the way. You could blend billboards with real shapes on a LOD basis so that you get real shapes close-up and shapes at a distance. You could even use imposters instead of billboards for the close-up work.

As you can imagine, it can get really complex and the trivial discussions I see here don't do the real task of foliage rendering any justice at all. My 'fxFoliageReplicator', although it works fairly well, is pretty simplistic and there are so many improvements that would need to be made to make it really efficient. Work has already been done with this with the fxGrassReplicator but there's still more work to do. These resources are a great basis for further work and some people have already done so.

I'd like to see me working on improving some of my existing resources (as time allows) to take advantage of batch vertex transfers, better LOD handling and more sophisticated culling. Another method would be to take advantage of the new TSE to accelerate and provide a feature-rich solution for generalised scene-clutter such as foliage, rocks and general debris.

Asking the question, "can torque do this" is kind of like asking "can a spoon cut fish". Sure, not as it stands but with the appropriate modifications, yes. ;)

- Melv.

#18 Also, the tree models supplied with Torque are ridicolously high polygon for their shapes IMO.

#19 @James: You made the correct purchase on the Quadro FX 3000; I'm using one and you'll have to trust me that I find it generally a higher quality card than the Wildcats. I think the OpenGL drivers from 3DLabs are top-notch, then again, they always have been. I've been through almost every generation of cards from them, even when the fill-rates were pretty low compared to todays standards, their vertex processors were always *fast*, especially when having dual-vertex processors was a radically new thing.

The 512Mb can be attractive for any volume work but I think we're going to see a nVidia 512Mb quadro-style card sometime soon. ;)

- Melv.

#20 No kidding. If I really was going to do this i'd make my own trees. I was only expermenting.

Rendering 10,000 objects on the screen is going to be hard to no matter what Optimizations you have. if you are 10,000 billboards that still 20,000 triangles.