Euclideon Island Demo 2011 "Unlimited Detail"

Toggle sidebar Toggle sidebar
T

T101

Senior member
Oct 13, 1999
558
0
76
I am not that surprised. Before when computers where slow (486) there was games that used block-like graphics to handle more detailed landscapes, because handling that many polygons was to slow at that detail level (some helicopter games etc. that I dont remember the name to anymore). Given the advancement in processorpower, I dont see why you could not do something similar these days. Of course there is some trick to getting the performance at that detail level, and I dont know how they do it. But still, it is interesting. We will see where it leads.
 
N

nonameo

Diamond Member
Mar 13, 2006
5,902
2
76
This reminds me of voxels from tiberian sun.

edit: oh, look at this... lots of talk about voxels :p
http://www.wired.com/gadgetlab/2010/04/unlimited-detail-3-d-graphics/

T101 said:
I am not that surprised. Before when computers where slow (486) there was games that used block-like graphics to handle more detailed landscapes, because handling that many polygons was to slow at that detail level (some helicopter games etc. that I dont remember the name to anymore). Given the advancement in processorpower, I dont see why you could not do something similar these days. Of course there is some trick to getting the performance at that detail level, and I dont know how they do it. But still, it is interesting. We will see where it leads.
Click to expand...

Perhaps it is prerendered :p

I won't believe it till someone provides an account of a live demo.
 
Last edited:
Phynaz

Phynaz

Lifer
Mar 13, 2006
10,140
819
126
Soccerman06 said:
If this was legit, wouldnt it be available to the public by now?
Click to expand...

First, this is for design, not rendering, so I don't know applicable to the public it could be.

Second, why do you say this? Do you think all technology has already been developed?
 
J

jimhsu

Senior member
Mar 22, 2009
705
0
76
The problem is exactly that their content right now is extremely repetitive. For example, compare the trees, dirt, layout of rocks, etc. If enough variation could be introduced (either by making more content - time and money; or procedural generation - hard and computationally intensive), would this algorithm still be feasible?

If I understand their tech right, the fundamental premise of their technology (searching through voxel space quickly) is fundamentally incompatible with increasing scene variation; in other words, as the amount of variation increases, the search space has to increase at least linearly (or most likely, at least O(N*log(N))). In other words, we either get a horribly slow engine, or a fast engine that creates the type of artificially hyperrealistic scenes shown in the demo. I don't see an easy way around that problem.
 
Last edited:
gorobei

gorobei

Diamond Member
Jan 7, 2007
4,103
1,607
136
we've discussed this before.
http://forums.anandtech.com/showthread.php?t=2066933
http://forums.anandtech.com/showthread.php?t=2057748

as i said in the locked thread, the math and principles are sound.

critical points:
1) these are not voxels or hyper voxels, even voxels get turned into polygons at some point in the render.
2) these are strictly point clouds. while memory intensive, if done right it can still be viable.
3) for static/non deformable scene objects this approach is fine. you might even be able to get leaves to sway a little. Animation deformers of any cloud mesh would be nightmareish.

the update shows that they have addressed some of the issues. ability to convert non-scanned meshes into cloud data, lighting and shadow improvement, general scene complexity.

there shouldnt be a memory issue as long as they are instancing each discrete element. as long as the catalog of elements is assembled in a smart enough way that you can build any object out of little repeated bits, the amount of point clouds could be workable.

such an approach might actually work better for physics destruction and terrain modification.

all deformable/animatable creatures or characters would have to be done in traditional mesh polygons, but if you use a mix of both point clouds and mesh in a game the effect could be very impressive.

game class lighting will be the next hurdle as the number of dynamic lighting effects in games is currently pretty high and without normal data the ability to light these point clouds could be limited.
 
SirPauly

SirPauly

Diamond Member
Apr 28, 2009
5,187
1
0
gorobei said:
3) for static/non deformable scene objects this approach is fine. you might even be able to get leaves to sway a little. Animation deformers of any cloud mesh would be nightmareish.
Click to expand...

I was curious about that based on how much things were static. The lack of dynamics and the lack of talk about this from the developer makes this a bit disingenuous.
 
I

Idontcare

Elite Member
Oct 10, 1999
21,110
64
91
I thought the details of the base of the tree were rather striking.

Unlike the hype that went into Lucid's "perfect scaling" claims, the claims presented by the narrator in the video are at least technically possible.

But as others have rightly pointed out, just because it is possible does not mean it is desirable when an actual gaming environment is involved.

Personally I enjoyed the video, progress in any industry goes through phases that are linear and predictable at times (hard-drive progression in areal density) and time which are quite non-linear and unpredictable (transition to SSD's).

Sounds like the key enabler for this technology is the availability of gigabytes of fast/cheap/ubiquitous ram. Something that wasn't around 10yrs ago and will only be cheaper and in larger densities 10yrs from now. (just like the enable of SSD's)
 
D

darkswordsman17

Lifer
Mar 11, 2004
23,444
5,852
146
BFG10K said:
Also no mention of how it scales with AF/AA.
Click to expand...

AA I would think wouldn't be as important as this is kinda like vector graphics in that the dot system should enable smoother and rounder objects to begin with.

Its interesting, but I think there's enough flaws that make it not particularly suited for gaming. It does sound similar to tessellation some, so maybe it can be used similarly.

I wonder if this might could function like pre-rendering? It would be pretty similar, low interactivity, but have some improvements, namely no fixed camera, plus it could enable random generation.

Its interesting, and shows the promise of what a full on point setup could enable in the future. Imagine pairing that with ray tracing, but I can't even imagine the processing power needed for that.

At the same time, I think that highlights that cloud computing might be necessary for gaming in the future, where instead of thousands or millions of computers processing similar content on their own, have supercomputers doing it. Likewise, it would be huge for games that are about interacting in a single world. Think of Minecraft on a supercomputer.
 
J

jimhsu

Senior member
Mar 22, 2009
705
0
76
SirPauly said:
I was curious about that based on how much things were static. The lack of dynamics and the lack of talk about this from the developer makes this a bit disingenuous.
Click to expand...

A related issue: how will collision be handled? Will collision meshes from the voxel data interact properly with player/monster models (such as a player leaning on a tree, for instance)? How much simplification do those collision models need?
 
J

jimhsu

Senior member
Mar 22, 2009
705
0
76
OK, thought experiment. Say that I took the tree scenes illustrated in the demo. Would the engine maintain performance without increasing memory/storage requirements if I:

1. Changed the X,Y,Z positions and rotations randomly for every leaf on every tree in the scene?
2. Perturbed the position data for every "vertex" of every leaf randomly by 0.01 (again, X,Y,Z)?
3. Procedurally generated unique texture "content" (I'm not sure what the voxel equivalent is) for every leaf on every tree?
 
gorobei

gorobei

Diamond Member
Jan 7, 2007
4,103
1,607
136
These are NOT voxels. Please dont use the term if you dont understand what is going on here. Everytime you use the term you are calling in a dozen different connotations with past voxel implementations that have nothing to do with this, and generally sidetracks the thread with people waxing sarcastic about some past game's crappy graphics.


High resolution point clouds are vertices and nothing else. there is a color component and possibly a specular value, but nothing else.

Voxels are 3d building blocks, but eventually get converted to polygons and then rasterized to pixels.

Point clouds have no polygons. The absence of polygons, edge data, surface normal, polygon addressing, and triangle setup is where the perfomance increase is coming from.
 
I

Idontcare

Elite Member
Oct 10, 1999
21,110
64
91
There are some good questions in this thread, you guys should email them and ask them directly. Would be awesome if you convinced them to join the forums and answer questions directly in this thread:
Please direct all inquiries to:
info@euclideon.com
Click to expand...
 
N

nonameo

Diamond Member
Mar 13, 2006
5,902
2
76
gorobei said:
These are NOT voxels. Please dont use the term if you dont understand what is going on here. Everytime you use the term you are calling in a dozen different connotations with past voxel implementations that have nothing to do with this, and generally sidetracks the thread with people waxing sarcastic about some past game's crappy graphics.


High resolution point clouds are vertices and nothing else. there is a color component and possibly a specular value, but nothing else.

Voxels are 3d building blocks, but eventually get converted to polygons and then rasterized to pixels.

Point clouds have no polygons. The absence of polygons, edge data, surface normal, polygon addressing, and triangle setup is where the perfomance increase is coming from.
Click to expand...

I understand I got dangerously close to calling them voxels but... technically I kept my foot of that one, final tile.

:)
 
B

bunnyfubbles

Lifer
Sep 3, 2001
12,248
3
0
gorobei said:
we've discussed this before.
http://forums.anandtech.com/showthread.php?t=2066933
http://forums.anandtech.com/showthread.php?t=2057748

as i said in the locked thread, the math and principles are sound.

critical points:
1) these are not voxels or hyper voxels, even voxels get turned into polygons at some point in the render.
2) these are strictly point clouds. while memory intensive, if done right it can still be viable.
3) for static/non deformable scene objects this approach is fine. you might even be able to get leaves to sway a little. Animation deformers of any cloud mesh would be nightmareish.

the update shows that they have addressed some of the issues. ability to convert non-scanned meshes into cloud data, lighting and shadow improvement, general scene complexity.

there shouldnt be a memory issue as long as they are instancing each discrete element. as long as the catalog of elements is assembled in a smart enough way that you can build any object out of little repeated bits, the amount of point clouds could be workable.

such an approach might actually work better for physics destruction and terrain modification.

all deformable/animatable creatures or characters would have to be done in traditional mesh polygons, but if you use a mix of both point clouds and mesh in a game the effect could be very impressive.

game class lighting will be the next hurdle as the number of dynamic lighting effects in games is currently pretty high and without normal data the ability to light these point clouds could be limited.
Click to expand...

yeah, he kept mentioned how detailed the dirt was, and all the while I kept thinking "good luck actually blowing that dirt around"
 
Ghiedo27

Ghiedo27

Senior member
Mar 9, 2011
403
0
0
Hypothetically, I wonder how feasible it would be to implement cloud point environmental detail the way that they used to have pre-rendered backgrounds (like FF7). The narrator mentions that the developers write for polygons which are then converted. Perhaps there could be an on-off switch for what gets converted.

If cloud point rendering has such great processor savings, then you could have a huge leap in detail by upping polygon count for animate objects. Bundle it with proper physics support in DX18. :D
 
gorobei

gorobei

Diamond Member
Jan 7, 2007
4,103
1,607
136
nonameo said:
I understand I got dangerously close to calling them voxels but... technically I kept my foot of that one, final tile.

:)
Click to expand...

lies, you took your hand off the rook. the move was final. :p
(actually wasnt directed at you.)

The quickest, most practical application of this would be a flight sim game.
-Full detail terrain that you dont get too close to (unless you crash),
-no model pop up at lod/distance change,
-no major deformations,
-pretty simple lighting (barring ground explosions of ground vehicles/targets).

add in traditional mesh enemy fighters and the visuals would be very impressive.
 
M

Mopetar

Diamond Member
Jan 31, 2011
8,525
7,785
136
Right now it's really only feasible for static elements, such as level design. Anything that requires a lot of constant motion isn't going to work. I'm assuming that they're working on tackling this problem and will probably start releasing more stuff once they have a better handle on it. My best guess is that they'll just pre-determine the possible animations and cycle through the different point cloud arrangements rather than trying to actually move the individual points.

Other than that, there aren't any real reasons why it shouldn't work. It would be interesting to see how well a hybrid solution would work.
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom