Procedural stuff benchmak

Hi,
do you remember this?
I worked again on it and moved most of the work on the gpu.

Would be cool if could download this small benchmark and post the calculation times for the different calculation modis and you pc spec.
DOWNLOAD
simply press 'c' to create a new tower, write down that calculation time, press 'm', press 'c' again, and so on.
Would be really cool. Btw, this is optimized for DualCore Cpus but it works also on single cores, and please close alls background programs to get correct numbers.
Oh and you need a graphic card with shadermodel3.0 support (Geforce6, Radeon X1000 and up).
A small screenshot to make you wet:


To make this post looking even more professional and scientific a pie chart for you:


[edit]
I forgot to mention you need the latest directx dlls (november2008?), you can find it somewhere on microsoft.com.
[/edit]
greetings
ChrisB

That seems to be some very well commented code :P

For those it doesn´t start, download the latest dx version here: http://www.microsoft.com/downloads/detai...;DisplayLang=en

cpu + density map on gpu: 23,675sec the result runs with about 55fps
gpu: 3,211sec but there are several faces missing and it runs with "only" 20fps
gpu + weld on cpu: 54,041sec runs also with about 20fps and is also missing faces...

It looks very cool :P
I am just missing some blur and disortion of the "clouds"

oh - this seems very cool
with cpu +density i get 5,151sec (and ca 70 fps)
with gpu 1,32 sec -> but only 20 fps
and a image like this:
same with gpu+weld on cpu (there it needs 12,185 sec ; 28 fps)

edit: o my specs are: Core 2 Duo E8500 3.8 Ghz (overclocked)
4gig RAM
GF 8800 GTS 512 MB (overoverclocked - with special coolingsystem XD )

Thx so far, there seems to be a problem with some gpus.
Please don't forget to mention your pc specifications.

That looks fantastic! Almost a game: Try to get up this tower, and find secret caves where you put items and hints for other gamers...

EDIT: Can't be of help, because my grafikcard doesn't support shader 3.0.

CPU Intel Q9300 QuadCore @ 2.50GhZ
GPU ATi Radeon HD 3870 X2

gpu
~25fps
2.039 seconds
Clutters

gpu + weld on cpu
30fps
3.716 seconds
Clutters

cpu + density map on gpu
89fps
6.205 seconds
No Clutters

These aren't the most exact as I have some background programs open.

I love this kind of stuff! Mathematical toys

You had to do a lot of things directly with directx, right? because 3dgs doesn't provide tools for actually create geometry.

Having same visual display as Revo
I see this on,
gpu + weld on cpu
gpu
Cpu: intel q6600 2.4 ghz,
Gpu: ATi HD3870

cpu + density map on gpu
127 fps
10.356 seconds

gpu + weld on cpu
28 fps
4,089 seconds calc

gpu
22 fps
1.041 sec calculation time


and as Inestical
but running plenty of background programs

Nvidia 9600GT (runing on Win7 Beta Drivers - i will switch to newest vista drivers and give it another try)

Q9300 quadcore at @ 2.50

screens:


EDIT: apperentally, i was not on latest win7 drivers, i will try with latest win7 and vista drivers

same problem as quadraxas
cpu + density map 9.181 secs >200fps
gpu 1.793 secs 60fps
gpu + weld on cpu 19.6 secs 40fps
EDIT
core 2 duo e6850
geforce 8800gt 1gb vram
4gb ram
win xp

CPU + Density map on GPU: 17.228 s @ 80 fps
GPU: 1.523 s @ 24 fps (the distortion problems like the others)
GPU + Weld on CPU: 23.034 s @ 26 FPS (same here)

AMD Athlon64 X2 6000+
nVIDIA GeForce 9800GTX+
2 GB RAM
Win XP 32Bit

CPU=6.375 sec/154 fps while looking at tower(300+ elsewhere) fps
GPU=2.094 sec/75 fps (same distortion)
GPU & WELD 11.3 sec/85 fps (same distortion)

vista 64bit
ultra 790i nvidia chipset
4 gig 2000 mhz ddr3 memory
twin 9800 gx2 geforce SLI
quad core 3.0 ghz Pentium
dx 10

edit:my cards is supposed to support shader 4.0...so I guess I am amzaed that I am getting these distortions also..

I also had firefox and steam running in the background so it might have mad a difference.

ChrisB
Your goal is to create the same effect as the from NVIDIA Cascades demo?
It would be nice to see the shader effects on it, but anyway would be very impressive project.



Look at this video and Please wait a little.
http://www.public.gamefactor.eu/towertest.swf

My specs:

Operating System
Windows System : Microsoft Windows Vista Ultimate 6.00.6001 (Service Pack 1)
Platform Compliance : x86

Processor
Model : Intel(R) Core(TM)2 Duo CPU E8400 @ 3.00GHz
Speed : 3GHz
Cores per Processor : 2 Unit(s)
Threads per Core : 1 Unit(s)
Type : Dual-Core
Internal Data Cache : 2x 32kB, Synchronous, Write-Thru, 8-way, 64 byte line size
L2 On-board Cache : 6MB, ECC, Synchronous, ATC, 24-way, 64 byte line size, 2 threads sharing

System
System : Gigabyte Technology Co., Ltd. EP35-DS3P
Mainboard : Gigabyte Technology Co., Ltd. EP35-DS3P
Bus(es) : PCI PCIe IMB USB FireWire/1394 i2c/SMBus
Multi-Processor (MP) Support : Yes
Multi-Processor Advanced PIC (APIC) : Yes
System BIOS : Award Software International, Inc. F2
Total Memory : 4GB DIMM DDR2 DDR3

Chipset
Model : Giga-Byte P35/G33/G31 Processor to I/O Controller
Front Side Bus Speed : 4x 333MHz (1.33GHz)
Total Memory : 4GB DIMM DDR2 DDR3
Memory Bus Speed : 2x 400MHz (800MHz)

Memory Module(s)
Memory Module : Corsair CM2X2048-6400C5DHX 2GB DIMM DDR2 PC2-6400U DDR2-800 (5.0-5-5-18 3-22-6-3)
Memory Module : Corsair CM2X2048-6400C5DHX 2GB DIMM DDR2 PC2-6400U DDR2-800 (5.0-5-5-18 3-22-6-3)

Video System
Adapter : NVIDIA GeForce 9800 GTX/9800 GTX+ (512MB DDR3, 625MHz/1.45GHz/2x1.10GHz, PCIe 1.00 x16, SM4.0)

Graphics Processor
Adapter : GeForce 9800 GTX/9800 GTX+ (1.01, 512MB, 1.69GHz)

Thx everybody for testing it.
HERE is a new, fixed version for you.
Slin helped me alot with testing, thank you Slin.
The new version should run on all ShaderModel3.0 cards. The gpu+weld on cpu mode should also be way faster now, so please give it a try again and post your results.

Thx
ChrisB

my numbers:
cpu: 24.5sec
gpu: 13.5sec
gpu+weld: 16.5sec

PC:
PentiumD 820 (2*2.8Ghz)
Geforce 6600GT
1GB ram

you fixed some of the rendering problem for gpu and gpu+weld on cpu but there is still some left


gpu = 28fps 2.131 sec
its ragged| not smoothed and lots of holes in the generated object.

gpu,weld on cpu
110fps
2.221 sec calculation
its not ragged but lots of holes in the generated object.

cpu density map on gpu
8.543 calculation
fps = 100-99

cpu: 7.23
gpu: 2.22
gpu + weld on cpu : 3.69

gpu one is not welded and not smooth.


9600gt sonic
core2quad q9300 @ 2.5

New tests with the "fixed" version:

CPU + Density map on GPU: 15.870 s @ 83 fps
GPU: 1.726 s @ 10-32 fps (jumping fps, unsmooth-shading)
GPU + Weld on CPU: 4.818 s @ 102 FPS (A LOT IMPROVED, some shading abnormalities)

AMD Athlon64 X2 6000+
nVIDIA GeForce 9800GTX+
2 GB RAM
Win XP 32Bit

WinVista Ultimate SP1
2 GB RAM
Core 2 duo 2.53~
Nvidia GeForce 8600 GT

GPU: 4.458sec @ 17 FPS
GPU + Weld on CPU: 4.458sec @ 17 FPS (Duh! The same!)
CPU: 4.458sec @ 17 FPS

Hi,
it took me the whole weekend to record this two little videos, i hope you like them.

It uses Newton for physics. Sadly building one tower takes nearly 10min, mainly because i didn'T figured out how to use newton in several threads and those pipes need really long to calculate.



Watch in HD!

upcoming contribution: ingame video recorder. Save your game scene directly as .avi or ogg theora, without any extra tools.

That's looking great.
I'm always thinking of water rushing down on this rock. Waterfalls and such. I think it would fit the scene very good.

The contribution sounds interesting, is it going to save every calculated frame then?

Love your tower concept and the shading looks impressive.
The ingame recorder is interresting as making movies is a long drawn out process.

I really like this project. don't stop!
I downloaded the demo as well, it's very cool. especially on the structure + texture side... it's such a complex structure, but I still can't see any seams... very cool.
What would be nice would be a demo with more adjustable parameters. i have no idea how all this works, but you do have some sort of primitve or sets of parameters that make sure the path is always correctly formed, right? if the user could adjust that, I'd really like it

Hi, its me again.
I worked a bit on some volumetric clouds for the game, they are completely dynamic. I think they already look good enough to show them, though there are still many issues, like the shading which must be better (currently its just tracing upwards, if it hits another cloud it get shaded darker). (i know there are some sorting errors, but i will not change it as this would require to much cpu power)

The clouds consists of several thousands of billboards, which is not that a big problem itself but because of the alphablending it really hurts the fillrate (i hope i can solve this with rendering the clouds to a smaller rendertarget seperatly).

Some screenshots: (note, the clouds doesnt get counted in the vertexes and polygon numbers)
taken on a PentiumD 820, Geforce6600GT


using a different texture:

more screenshots on my website.

Finally something for the scientist of you:
For simulating a cloud physical correct you have to calculate every 0.01mm thick waterdrop. Because this would be to much work i don't do that. My method is way faster than the physical approach.
In numbers:

(Don't take the numbers to serious)

greets
Chrisb

lol

im so in love with your tower concept I have to make one like it

Do you use small sprites with roll to layer your clouds?

Would small animated ones drag too much do you think?

Hi, just a small update this time, mainly a tribute to the 4gb ram upgrade.
I'm currently trying to increase the "omg, this is huge"-feeling and added a simple landscape and made the tower 5 times larger. It is as detailed as before but uses 4 lod steps.
The voxel map of the tower is 256*256*2560 big.



more screenshots
Next task is to add some sort of atmospheric scattering to the shading to get rid of the simple blue fog.

I've also added a .wmb level output (a7.7 format, works with lite-c), though lite-c seems not to be able to load levels with millions of triangles. Here some screenshots of a smaller tower in lite-c:



(lite-c is cheating on the number of tris in the depug display, the level consist of twice as much) (its also possible to export lod-meshes which are much more optimized)
You can also download the level: level_1.rar
simply open it with acknex.exe, or load it with a script.


As the geometrie gets calculated via several shaders i originally planed to take part at the shader contest. Though i think i won't get many votes with this as it is not a shader to create fancy looking surfaces or a complete easy to include shader package. For most of the user its simply useless.

omg, this is huge... :P

nice work... I'd like to see more rounded edges at the top though -> the transition from rock to grass is much too sudden in my oppinion.
Also, when will you make a game of this?

Chicken chicken chicken chicken chicken chicken Chicken chicken chicken chicken.

Chicken chicken!
Chicken chicken chicken chicken chicken chicken chicken chicken chicken chicken chicken chicken chicken chicken chicken chicken, chicken chicken chicken chicken Chicken chicken chicken chicken chicken. Chicken chicken chicken chicken chicken chicken chicken chicken chicken! 30 chicken chicken chicken chicken chicken chicken chicken 0.5-1 chicken chicken chicken chicken! Chicken chicken chicken!

Chicken chicken chicken chicken chicken? Chicken chicken?

chicken? oh YEAH CHICKEN !!!!
Great Stuff!

cYa Sebastian

Totally love it!

Still I would like to see this used in a game. Procedural asteroids is something I could really use for my project atm. Or a procedural world, maybe with less detail. Or a procedural mountain with caves in it...

Of course, I have no idea about this stuff, so I don't know if any of the above is realistic.
Are you going to make a game with it? or are you just messing around for fun atm?

Great progress, the lod-steps seem to work real nice...

This is simply brilliant!

Just an idea for this as part of a game:
Place at top of the model a sphere that the player has to lead a way down to a certain goal at the ground of the model. Give the player a set of objects that he can place on the model. When he clicks start, enable the physics for the sphere, and he can watch the sphere rolling down and whether his objects are well placed to lead the sphere its way to the goal.

Thx for the kind words.
Yeah i'm still not sure how to make a game out of it. Any ideas are welcome. This scene was more or less for fun. I would like to explore such worlds in a rpg, but then again a rpg is sooo much work.

A weird shooter with exaggerated fancy effects would rock

Yeah, a flying shooter, with jet packs, a sort of capture the flag. With 'A' and 'D' you move right and left of the tower, with 'W' and 'S' you move up and down.
As a condition you need something that gives you reason to hide behind the obstacles.

a plane shooter would also fit the scenario, make too teams wich start from two different platforms with their planes, add a good working flight control mechanism (love the one of Battlefield Heroes ;)) and you got your game

I thought of a space glider

Yeah, you need to fly from time to time through the tower to get a lift by the warm air. Depending from where the sun shines on the rock, you have different places and conditions to gain height.

Maybe, you'd like to start a simple multiplayer, where everyone can simply fly and walk around. After that you can add experimental features, so that the community can test whether they are fun or not, and give further suggestions.

Just make a cinematic trailer .

i would just remake starfox64's flight mechanics and use these as levels and make it multiplayer, cant beat that really

if this world were darkinferno's, all games would be multiplayer games

No, I agree though. Flight simulator of some sort would be cool, but I'd add a few of your towers, not just one. (Even if you don't make a game out of it, can you make a video with mutliple, maybe a little smaller, towers? I think it would look real fun :D)

Also, i'm still interested in what sorts of shapes this code can produce. You've only shown towers so far, which look awesome, but could it also create rocks/spheres/hills/caves? Is there any input-parameters that are used to control what sort of shape it is (I'm sure there is, but how "flexible" are they?)

Reason I'm asking is: we want/need big asteroid-fields in our game. And we simply can't get rid of the seams (in the modeling tools they look great, but as soon as we import them to acknex, there's the seam), and we needs lots of different asts, so we were thinking of using "procedural stuff" for the larger ones.
Any insiders as to whether this would even make sense at all? With your LOD code, it would truly rock...

EVE online has done something like that, though I don't know if they're using procedural shaders:
http://ccp.vo.llnwd.net/o2/devblog/img/DevBlog_WIP_02_s3.jpg
Maybe I should read up on that some time...

Do you use any kind of relief mapping?

I think a "glider-like" game would be really cool; 360° action with dogfights...

ok, hand over the Glider core and we'll set it up

j/k

Hm i like the idea of a glider game . Maybe with something like the glider from assassin creed 2 http://i.computer-bild.de/imgs/90731031_2eec319d3f.jpg

You can create any shape you want, you can find some older screenshots on my website http://www.swollen-eyeballs.org/site/pages/experiments/procedural-terrain.php?gallery_id=16
An asteroid should be easy, first add a sphere, add several layers of noise and then remove some smaller spheres to create craters.


Nope, but some normalmapping in the las video.

Don't forget LOST

An asteroid with lots of tunnels and holes would indeed rock

Well I would think of a worms-like game with procedural terrain! This would really benefit from that. And ifd it even is possible to shoot away the terrain it'd be great X-D

Screenshots of some asteroids:
http://www.swollen-eyeballs.org/zeugs/ast/index.php
Keep in mind: You're loosing your insurance coverage when flying through an asteroids field!

looking great! 22 fps just won't be enough for a good game... and not when I want around 100 asteroids showing at a time... We wouldn't need the asteroids as detailed as they are here though. So maybe it would work with lower LOD-steps...