RumbleSushi 3D game demos

Hello,

I've known about Flashkit for years, but only recently did I even realise it had a forum ;)

Anyway, I thought this would be a good place to post some demos built with my proprietary 3D engine, RumbleSushi3D.

The engine has been built from the ground up, purely for making games. So as you can imagine, I've worked hardcore on performance.

The engine is almost finished, I just need to import 3D model animation, implement some fast 3D collision detection, then it's on to actually making games. I'll start a blog when games start production, giving info about upcoming games, and possibly some playable demos etc too.

Feedback is welcome, I'd be especially interested in what framerates you're getting on what CPUs.

Here is my latest demo. 150 buildings, 500 moving flowers, mip mapping, frustum clipping, perspective correction - running at 60fps on a Core 2 Duo PC.

http://rumblesushi.com/city.html

I'll be happy to post some earlier demos too, if requested.

Cheers,
RumbleSushi

I forgot to add - you use the mouse to look/turn, and move with the arrow keys or WASD. If you want to enable vertical look, press V.

Cheers.

Runs fine, about 30 fps on my nVidia Ion PC.

30fps or above is good, but it's strange some of the framerates people are reporting.

I've tested this demo on an ancient AMD Sempron, 1.6 ghz, with 512 megs of RAM, and it runs at a near constant 60fps, with the occasional dip to 50 something.

The CPU in your netbook is almost as fast as a 1.6 ghz AMD Sempron, I would have thought you'd get around 50fps at least. Unless you're using Chrome.

Man!!! When I'm used to getting about 5-15 fps in 3d demo's, you're engine actually ran 24-30 fps, though 28 most of the time! It looked really smooth. I give you a mega thumbs up in performance (would you mind giving some background details on the techniques you used? There are some people on this forum that are devoted to 3d in Flash!)!

Looks really really good! I hope I can use it someday...I never have time to finish my engine...

P.

I oscillated between 60 and 61 fps the whole time, great job !
Can you tell us more about it ? Developing it must have been interesting :P

Well, runs fine at 60 fps over here, with a pretty standard computer I think
Only thing I noticed off was some weird fluctuations in the image, most particularly around the edges of the screen

Eh?
Otherwise, good looking city me thinks

Excellent, glad you like it.

Pazil - what CPU do you have?

Developing it is very fun indeed, especially seeing as I've done it alone, it's been quite challenging.

As I mentioned in the first post, what I'm working on now is proper 3D per polygon collision detection. Which doesn't seem particularly easy, especially getting it run without having a noticeable effect on performance. The collision detection in the demo above is just a quick and dirty 2D collision, seeing as the player's Y coordinate doesn't change, and the objects to collide with are just rectangles.

The techniques I've used aren't particularly revolutionary. Just some clever code, a very streamlined overall pipeline - very quick per model transformations, a quick polygon pipeline, and a decent amount of deferred rendering/deferred calculations. I've tried to keep things as simple as possible. The engine is not bloated or overcomplicated in the slightest. I've kept classes to a minimum, and always taken the quickest route from A to B.

And because it's geared 100% towards game development (I don't plan on making websites, unless I get a particularly good offer) - my focus has been on speed, speed, speed from the get go. I've benchmarked all areas of the engine. There's been a somewhat obsessive amount of benchmarking, which has been necessary to get it running as fast as it does.

One of the reasons I've focused on quick model handling, is it goes without saying that you're not going to have a particularly realistic high poly 3D model in Flash - with no GPU, your polygon budget is obscenely low.

So even with just 1 model using 100% of the polygon budget, the result is still going to be a low polygon model, by console standards ;)

So I think it's more impressive to say have a large amount of really simple enemies or objects, rather than just a handful of slightly less simple enemies/objects ;)

But enemies aside, the faster the per model handling, the more background objects I can have, as long as there is some clever level design, not allowing too much to be in view at any 1 time.

I'll post some earlier demos tomorrow, demonstrating some very quick flock code, and a slick tracking cam I developed.

One of the demos has 1000 simple flat shaded, moving spaceships (just 6 polygons each) running at 30fps on a Core 2 Duo.

The demo I posted above was actually something new for my engine, creating a larger scene, with larger objects etc, and new features like mip mapping, perspective correction etc.

Anyway, I'll post the other demos tomorrow ;)

And the demo I'm working on now is just a practice run for 3D per polygon collision, it's some basic snowy terrain, with trees etc.

Cheers,
RumbleSushi

Trogdor - that's a side effect of dynamic triangulation, which is used to correct the texture warping of affine texture mapping. You get broken textures and texture swimming.

To be honest, for most busy/messy textures, dynamic triangulation is fine, but with those really clean grid textures on the sides of the buildings, it really shows the flaws of the technique.

This demo should be more to your tastes, the Flash 10 version. Completely different rendering technique, it uses more juice, but is pixel perfect.

http://rumblesushi.com/city_F10.html

If anyone wants to know more about how dynamic triangulation and the Flash 10 technique work, let me know, I wrote a post about them earlier.

Quote:

---

Originally Posted by rumblesushi

30fps or above is good, but it's strange some of the framerates people are reporting.

I've tested this demo on an ancient AMD Sempron, 1.6 ghz, with 512 megs of RAM, and it runs at a near constant 60fps, with the occasional dip to 50 something.

The CPU in your netbook is almost as fast as a 1.6 ghz AMD Sempron, I would have thought you'd get around 50fps at least. Unless you're using Chrome.

---

I tested on Opera and IE, Win7. fps remains around 30 so it could be something about that Intel Atom CPU.

Thanks for the info tonypa. Not unlike PC games, windows/flash as a platform can be annoying in that user's systems are unpredictable, and it's hard to know what level of performance to target/expect. Either way, I guess 30fps from a proper 3D scene like this running on a netbook, can't complain.

Anyway - here is my first demo that experimented with moving objects, this is from November. 100 textured flowers in an undulating line flock.

Press C to change the camera to a rollercoaster cam. Press C again to change it back. By default the camera is a sort of tracking cam/hover cam, much like a replay camera in a racing game.

Here it is - http://rumblesushi.com/flock_01.html

gotta say, this is looking very nice. And thank you for thinking about a 3d engine for game development, low poly, lots of models.

This demo has models galore. It's a flock of uber simple spaceships, just 6 polygons each. But they flock and there 360 of them, running at 60fps on a Core Duo or above.

Press D to see some debug stats, and press M to load in another 640 ships, bringing the total to 1000. The framerate switches to 30fps with this mode.

http://rumblesushi.com/flock_02.html

Cheers,
RumbleSushi.

I really really want you to know that I love your engine!

Seriously, it's the only one giving MY computer "good" performance! With 360 models, I was running it at 18 fps, when others will crawl with that. When I went to 1000, then I could really see a hit to performance, but still...if it was a game, it was moderately playable still.

I have a AMD Athlon @ 1.15 Ghz, and 768 Mb of RAM.

It seems that you took the path I was starting to take, doing manual bitmapData drawing stuff, rather than the slow slow built-in vector drawing...In fact, I'm still working on a z-buffer concerning that...an on-and-off project.

I'm really interested to see more demo's and hopefully finally some source!

P.

I get around 60 fps and it runs smoothly but the key controls don´t work for me on safari with latest flash player. Actually the latest flash player is quite buggy for me for various things, maybe its that.
I like the demo itself (i could also run around with older player on my pc), its cool, sadly i have to say for flash.
I wonder:
-Why have you chosen to create your own engine instead of using away3d or one of the others?
-Why have you chosen to do realtime 3d in flash instead of using unity3d?
I don´t want this to turn into the same old unity vs flash discussion, i´d just like to get your view on it, i wonder that since it seems you want this to have game focus and games usually push the technical edge of flash´s 3d "capabilities" most, wouldn´t it make most sense for 3d games to be done in other tech?
I still sorta get it when advertising clients in some cases insist on using flash or a flash developer wants to stick with what he already knows and picks and uses one of the readymade engines and lives with the restrictions, but when i see a 3d engine developer like you i wonder especially what´s the motivation to do it in flash =)
Is it mostly that you want to bend the boundaries of what´s doable in flash or do you see other benefits in using flash for this type of stuff other than the high plugin penetration?
It makes me think more these days when i see something like this while on the other hand even the core team that started papervision3d initially has switched over to using unity3d mostly.

60 fps

3.2 ghz quadcore, Windows 7, ie8 and Firefox 3.5 (same frame rate in both)

ATI HD 5700 series card

Really impressive work mate, runs silky smooth for me on my mac book pro, the FP10 version of the original demo is excellent.

It's been a while since any 3D in Flash has been interesting, and this has really hit the spot.

Squize.

Thanks Pazil ;)

I'm actually using the graphics class to draw the polygons, bitmapFill.

Good to see you get a somewhat acceptable framerate on such an old machine too.

Tom - the reason for choosing to develop my own engine rather than using Away 3D or Papervision is quite simple. Partly for the challenge, but mainly because they are too slow to make proper 3D games. With any fairly complex scene with a lot of models, there's no way you're going to get 60fps out of Away3D or PV3D.

Obviously they are far more feature rich than my engine, but most of the features are more geared towards website stuff rather than games.

My engine has been built for speed, rather than usability or versatility, and the features are far more game focused - I'm focusing on LODing, mip mapping, enemy movement, collision detection etc.

Also, what better way of getting a thorough understanding of 3D mechanics than building your own engine? Building my engine and knowing how it works inside out, automatically puts me at a big advantage when developing 3D work, as I know exactly how everything is working, and can easily modify/customise things as I see fit.

The difference in speed alone is worth it, the other Flash 3D engines aren't fast enough to render a proper 3D scene with a lot of objects, movement, collision detection etc at an acceptable framerate.

As for your second question - penetration. From a business standpoint, 3D games in Flash make more sense, and there is a gap in the market for them. Despite technology like Unity being out there, people are going to be excited about playing an N64 or DS level game in Flash.

And I guess what it boils down to is 99% of users have Flash 9, and even Flash 10 is up to around 90% now.

Do you know the stats for Unity?

And I guess, being that these are browser based games we're talking about, I'm going to create some fast, fun, arcade style games. With a big pick up and play appeal, and score attack appeal etc.

I'm not aiming to make games that compete with console games. And in regards to the huge horsepower difference between my 3D engine and Unity (as Unity uses the GPU obviously, the difference in poly pushing power is enormous) - the DS Vs the PSP perfectly illustrates that horsepower only counts for so much.

The DS is more accessible, has a lot of fun games, makes the most of it's low polygon budget, and has very playable games at high framerates (most DS games run at 60fps, even Mario Kart DS).

So for the kind of fast, stylistic arcade style games I'll be making, I don't even really need the polygon budget of Unity.

With such a small Polygon budget too, it'll force me to be far more creative in regards to art style etc, coming up with interesting designs and textures that suit such a low polygon environment.

Don't get me wrong though, I find Unity 3D hugely impressive. Although I've known about it for ages, I only had a proper look recently, and I was blown away. I had a look at the Tropical Paradise demo.

I was expecting Shockwave level 3D visuals (which are impressive in their own right, as it is also hardware accelerated) - but the Unity visuals were more approaching PS2 level graphics. Very impressive.

Apart from the low penetration, one of the things that puts me off is, am I right in thinking it uses a scripting language, with no use of classes etc? I heard someone mention it.

Regardless, I shall give Unity a go once this project is properly up and running :)

Cheers,
RumbleSushi

A couple more demos. Here's an example of imported models.

http://rumblesushi.com/cars.html

Press the Space bar to swap the 2 cars with 20 really basic cars.

And here is an underwater scene I created, because I love fish - http://rumblesushi.com/sea.html

4 seperate shoals of fish, with their own characteristics.

Press M to load a Yann Tierson track to listen to while watching the fish ;)

Cheers,
RumbleSushi

Runs perfect here. 2.4GHZ Firefox 3.5 3MB DDR RAM.

Have to agree with trogdor though: The walls seem to be "waving" while I move, similar to what happened in the Matrix.

Other than that... nice work I guess.

ImprisonedPride - yep, that's a by product of dynamic triangulation, which is a homegrown method of correcting the texture distortion caused by affine texture mapping. In a nutshell.

I posted a Flash 10 version using a different rendering technique. It's pixel perfect but uses more juice.

http://rumblesushi.com/city_F10.html

The fish were beautiful! That reminded me of an old idea to make a 3d fish game, so...when do you expect to be finished the engine? (Sorry for being pushy... :) )

I like making a 3d engine in Flash a) it forces you to really work with optimization, and be creative in implementing certain stuff. I like to see good 3d games in Flash, and with this engine I can see some finally come! It's like rumblesushi said, it forces you to be creative also with what you design in the game and level planning!

Great engine! And post immediately as soon as it becomes available!

P.

Quote:

---

Originally Posted by rumblesushi

Thanks Pazil ;)

I'm actually using the graphics class to draw the polygons, bitmapFill.

Good to see you get a somewhat acceptable framerate on such an old machine too.

Tom - the reason for choosing to develop my own engine rather than using Away 3D or Papervision is quite simple. Partly for the challenge, but mainly because they are too slow to make proper 3D games. With any fairly complex scene with a lot of models, there's no way you're going to get 60fps out of Away3D or PV3D.

Obviously they are far more feature rich than my engine, but most of the features are more geared towards website stuff rather than games.

My engine has been built for speed, rather than usability or versatility, and the features are far more game focused - I'm focusing on LODing, mip mapping, enemy movement, collision detection etc.

Also, what better way of getting a thorough understanding of 3D mechanics than building your own engine? Building my engine and knowing how it works inside out, automatically puts me at a big advantage when developing 3D work, as I know exactly how everything is working, and can easily modify/customise things as I see fit.

The difference in speed alone is worth it, the other Flash 3D engines aren't fast enough to render a proper 3D scene with a lot of objects, movement, collision detection etc at an acceptable framerate.

As for your second question - penetration. From a business standpoint, 3D games in Flash make more sense, and there is a gap in the market for them. Despite technology like Unity being out there, people are going to be excited about playing an N64 or DS level game in Flash.

And I guess what it boils down to is 99% of users have Flash 9, and even Flash 10 is up to around 90% now.

Do you know the stats for Unity?

And I guess, being that these are browser based games we're talking about, I'm going to create some fast, fun, arcade style games. With a big pick up and play appeal, and score attack appeal etc.

I'm not aiming to make games that compete with console games. And in regards to the huge horsepower difference between my 3D engine and Unity (as Unity uses the GPU obviously, the difference in poly pushing power is enormous) - the DS Vs the PSP perfectly illustrates that horsepower only counts for so much.

The DS is more accessible, has a lot of fun games, makes the most of it's low polygon budget, and has very playable games at high framerates (most DS games run at 60fps, even Mario Kart DS).

So for the kind of fast, stylistic arcade style games I'll be making, I don't even really need the polygon budget of Unity.

With such a small Polygon budget too, it'll force me to be far more creative in regards to art style etc, coming up with interesting designs and textures that suit such a low polygon environment.

Don't get me wrong though, I find Unity 3D hugely impressive. Although I've known about it for ages, I only had a proper look recently, and I was blown away. I had a look at the Tropical Paradise demo.

I was expecting Shockwave level 3D visuals (which are impressive in their own right, as it is also hardware accelerated) - but the Unity visuals were more approaching PS2 level graphics. Very impressive.

Apart from the low penetration, one of the things that puts me off is, am I right in thinking it uses a scripting language, with no use of classes etc? I heard someone mention it.

Regardless, I shall give Unity a go once this project is properly up and running :)

Cheers,
RumbleSushi

---

Ok, i see regarding why you do your own one and go for using flash =)
I think your points are understandable for web deploy.
I started using flash back when the plugin was pretty new and not widespread at all yet, so yeah, i have no worries to use something on the web that hasn´t the most widespread plugin yet, especially for games where the audience and clients are more likely to adopt new tech faster when they see advantages in it.

Regarding unity plugin penetration here´s an interesting thread on that: http://forum.unity3d.com/viewtopic.php?t=41880

For me the deciding factors for using things like unity3d and other solutions more for the higher end or other deploy options than web playback stuff were mostly two things:
-Loosing the belief in Macromedia/ Adobe doing the changes and additions i think are required in any reasonable timeframe or at all
-The change in how i see myself regarding what i do: A few years back i´d see myself more in the hardcore core engine developer field. Sure, i wouldn´t code Assembler or similar low level stuff, but i´d find joy in spending lots of time in writing own physics engine setups or tweaking the performance side of game engines all day to squeeze out most of flash.
Nowadays i see myself more as game developer and game designer and that´s what i´d like to focus more on than a core engine coder.
I want to spend more time on making things fun than on making things run at all and also more on game logic and interaction than keeping the framerate up even for 2d stuff :)
So yeah, meanwhile for me the solutions that deliver the core functionalities out of the box and allow me to actually focus on making the content are meanwhile more interesting.

Also to me personally since i also want to deploy to desktops and other devices, too its more important that i can use something that leads to good compeitive results on all platforms and deploy ways, but yeah, to each his own =)


Regarding the unity island demo: Yeah, i agree, i was similarly blown away when i saw it the first time and yeah, well, its probably sorta more in ps2 level besides the anti aliasing and some shaders not possible on the ps2.
Regarding unity in general its also used for current gen console games, one can do a lot more than what would fly on the ps2 =)

Regarding unity only supporting scripting lamguages: That´s not right, i guess that misinformation is maybe founded on the point that one creates seperate files for the code and those are often referred to as script files.
But the actual content of those files, so the code can be done in Boo (only in web version) , Javascript and C#.
In Javascript one can use classes but doesn´t have to whereas C# is totally class based.

really nice demos you added meanwhile btw =)

60+ FPS in all demos, very nice.

(Core i7 920, Win 7 Pro, Chrome 3.0)

Quote:

---

Originally Posted by tomsamson

Ok, i see regarding why you do your own one and go for using flash =)
I think your points are understandable for web deploy.
I started using flash back when the plugin was pretty new and not widespread at all yet, so yeah, i have no worries to use something on the web that hasn´t the most widespread plugin yet, especially for games where the audience and clients are more likely to adopt new tech faster when they see advantages in it.

Regarding unity plugin penetration here´s an interesting thread on that: http://forum.unity3d.com/viewtopic.php?t=41880

For me the deciding factors for using things like unity3d and other solutions more for the higher end or other deploy options than web playback stuff were mostly two things:
-Loosing the belief in Macromedia/ Adobe doing the changes and additions i think are required in any reasonable timeframe or at all
-The change in how i see myself regarding what i do: A few years back i´d see myself more in the hardcore core engine developer field. Sure, i wouldn´t code Assembler or similar low level stuff, but i´d find joy in spending lots of time in writing own physics engine setups or tweaking the performance side of game engines all day to squeeze out most of flash.
Nowadays i see myself more as game developer and game designer and that´s what i´d like to focus more on than a core engine coder.
I want to spend more time on making things fun than on making things run at all and also more on game logic and interaction than keeping the framerate up even for 2d stuff :)
So yeah, meanwhile for me the solutions that deliver the core functionalities out of the box and allow me to actually focus on making the content are meanwhile more interesting.

Also to me personally since i also want to deploy to desktops and other devices, too its more important that i can use something that leads to good compeitive results on all platforms and deploy ways, but yeah, to each his own =)


Regarding the unity island demo: Yeah, i agree, i was similarly blown away when i saw it the first time and yeah, well, its probably sorta more in ps2 level besides the anti aliasing and some shaders not possible on the ps2.
Regarding unity in general its also used for current gen console games, one can do a lot more than what would fly on the ps2 =)

Regarding unity only supporting scripting lamguages: That´s not right, i guess that misinformation is maybe founded on the point that one creates seperate files for the code and those are often referred to as script files.
But the actual content of those files, so the code can be done in Boo (only in web version) , Javascript and C#.
In Javascript one can use classes but doesn´t have to whereas C# is totally class based.

---

Hey Tom, thanks for the compliment. Even though technically my latest demo is the best showcase, I still like those flock demos, and the underwater scene. The first two were from november, the second two from december. The latest demo was completed a couple of weeks ago.

Your reasoning is completely understandable too.

I must admit, I really do enjoy building this engine, and pushing AS3 for all it's worth. It's fun to me.

But at the same time, I've built a 3D engine that is perfectly catered to my needs, so when it's finished, I can then focus 100% on game development and design, just like you said. Which then wouldn't be that different from you using Unity, the difference is you obviously have to put in a LOT of hours to build your own engine rather than learn to use an out of the box package. So for most people, an out of the box package is a far more attractive option.

I guess for me though, this was actually necessary. It's fun, and challenging, but like I said, the main driving factor was that the out of the box 3D packages in Flash are too slow, so I had no choice but to make my own.

It's true what you say, gamers are far more likely to think nothing of installing a plug-in that takes like 30 seconds to install, compared to most people. However a large portion of the market are casual gamers, and I think it's more proper gamers that fall into the category you speak of.

I guess what it comes down to, is a good game published in Flash online, is going to be played by far, far more people than a good game in Unity online, right now.

That's not to say I plan on making ****, casual games. I'm going to make fun, accessible games that should appeal to both the hardcore and masses alike.

Thanks for the info regarding programming in Unity. That's good to know. I was under the wrong impression, because I did actually see someone complaining about programming in Unity, as if it was completely procedural and limited, like the Flash 5 revision of AS1 ;)

To be honest I've been restraining myself from trying Unity since I saw that Island demo. I'm 100% focused on this project now though, and I need to work on something everyday, or I lose my flow.

I was also quite excited to hear there is a free version now. The only problem is, the free version is lacking 2 features I consider essential, the low level access and profiler.

Just a couple more questions regarding Unity. Just how out of the box is it? I recall it uses PhysX right? So does it have physics/collision detection/cameras/model importers etc right out of the box? Is it that easy to get started?

And does the native physics engine/collision detection seem to run pretty fast?

Hey guys, just so we don't lose focus on rumblesushi's engine, I've forked the interesting discussion about competing 3D engines here:

http://board.flashkit.com/board/newt...=newthread&f=5

[edited by tomsamson] and i moved the posts in question =) [/edit]

rumblesushi, what format are you using to import things?Collada?

Incrue - right now, just OBJs. They are the easiest models to parse.

They are not that good though, I thought they would be lighter in filesize than colladas, but they aren't.

Before making games, importing Colladas or 3DS files with animation is near the top of the to do list.

I'll have a new demo to post soon. Proper 3D per polygon collision detection is a bit tricker than I thought. I need to get it working accurately with minimal impact on performance.

I'm also working on getting the Flash 10 rendering running faster.

Are you planning on implementing a scene graph, skeleton animation, and/or bsp or quad mapping to this engine?... then make it open source and I'll kiss ya add normal light mapping and pixel bender effects and i'll do more! lol

RipX

I once tried to get PixelBender filters to work...but I never did. PixelBender seems like a completely different planet to me from a Flash perspective. It would be cool to see how people got it to work, especially if you could implement it with a 3d engine!

P.

@Pazil Some flash 3d engines already use PixelBender to allow for shader effects on the GPU.

Really?! That's awesome!
Which ones?

I know that away3d has them for sure, but probably papervision & sandy also. Not sure about alternativa. Here is a few pixel bender implementaion examples: http://away3d.com/away3d-3-4-2-pixelbender-to-the-max

RipX

I'm not planning on implementing pixel bender filters or any dynamic lighting effects at all actually, unless I can come up with something that runs very, very fast.

As impressive as some of the pixel bender shaders/lighting effects are, implementing them in a 3D flash game would be insanity. You'd cut an already tiny polygon budget in half at best.

They make for interesting experiments etc, but they are only really suited to something very simple with just 1 model, like the examples above. It wouldn't be realistic to apply lighting effects/shaders to a 3D game in Flash.

Oh and in regards to animation/BSP etc.

Importing animation in 3D models is essential obviously. I'll use either colladas or 3DS files.

As for BSP, probably a bit unnecessary. I'm working on a grid or quad tree for collision detection. For levels that are as tall as they are long/wide, I'd probably have to make an octree.

And about lighting, I think for the most part baked textures look fine, as long as the graphic artist does a good job.

@rumblesushi - I agree on the points about pb filters for the most part, I think a serious look at BSPs might see a speed improvement for large models/areas and I think you should seriously look at it to allow your faces to be divided, this prevents the overlapping of faces that can create graphical glitches, would help with face culling etc and lod stuff might be good to look at if you want to get even more speed!

It would be nice to see where this goes! :) Keep it up!

RipX

Yep, that's what Alternativa mostly use the BSP tree for, as far as I know, very accurate z sorting, splitting/clipping polygons to other polygons, just like frustum clipping really, but with other polys.

However, this would undoubtedly be significantly slower than the good old painter's algorithm overall. To be honest I would rather have the odd bit of z fighting or poly overlapping, with maximum performance, rather than a lower poly count with uber accurate z sorting.

And I think with clever level design/modelling, you can eradicate most z fighting glitches.

I'd probably only consider a BSP for collision detection. But again, with the sort of polygon count we're talking about, I think it's best to keep things as simple as possible, I think for most levels a quad tree or even uniform grid would offer the best performance for per polygon collision detection (I could be wrong, as I don't know much about trees).

Someone else recommended me a BSP tree actually, and one of the things I overlooked, is if used only with static geometry, the tree would only have to be generated once, rather than every frame. However, at a guess, I think the increased look-ups on a complex BSP tree would probably outweigh the extra accuracy it would provide, performance wise.

The thing about pixelBender, is I think it's a common misconception that it uses the GPU for the rendering. The toolkit uses the GPU, the Flash player uses the CPU, just like it does for everything else. And it's insanely CPU heavy, just like any of the Flash pixel level filters.

Incidentally, you asked about open source - this engine is being developed for proprietary use, purely to make games, and that's one of the reasons it's so fast. I haven't had to factor in usability at all, I've just done everything as fast as I possibly can, at the expense of usability and doing things "properly".

So although I might change my mind, I don't plan on a public release. It would be too much work, I'd have to restructure the whole thing really, and it probably wouldn't run as fast as it does now ;)

PS - the city demo has mip mapping (texture LODing) - and for models with a decent polycount, I'll almost certainly implement geometry LODing too ;)

Cheers,
RumbleSushi

This looks good. I'm usually not big on demos, but it looks like the jump from where you are now to an actual game shouldn't be too hard and would not hinder performance much.

Runs at rock solid 60FPS on my 3 year old WPad.