Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Itnetlolor]

If this takes off as well as I hope it would, it ought to put the focus back more on gameplay since the graphics race will be limited to the artists of the games instead, so more funds will hopefully go towards game elements and other more important things that makes games good, instead of carbon-copying all sorts of things, slapping a different label on them, and calling it *NEW* or *BETTER GRAPHICS* by showing a CGI sequence instead of actual gameplay footage. That kind of crap is what ended up making, for example, Final Fantasy XIII such a disappointment. It was beautiful, yes; but it really lacked in gameplay and freedom. Neither owning a PS3 or 360, I can't really vouch for a review personally, but this is what I recall reading.

I do like the proposal of how it works. Let's hope it's not bullcrap.

I can imagine just how awesome it would be when running on hardware mode.

[Ampersand]

The thing is that it already looks awesome running in software mode, assuming they're telling the truth. The video did say they'd be releasing the SDK soon so it should be possible to take a looksie for ourselves once it's out.

[Armok]

I'm prety sure this is fake. I know a thing or two abaut computer grapics, and this simply is mathematicaly imposible no matter how clever algorithms you use. Even if it was posible, this kind of thing would be used by the movie industry long before it got remotely close to being able to run on a normal PC, and certainly not in realtime.

[Fooj]

It's a search algorithm for a point cloud that only grabs the points in space relevant to the camera. That much sounds plausible. What I don't understand is how the detail is "unlimited". It's not calculating surfaces inbetween points to make polygons, so all those points have to pre-exist. Every single pixel that could be rendered would need to be created and stored ahead of time. They would probably use high-poly models, texture them, then calculate a gigantic point-cloud file from that.

They would still need poly meshs for collision models. Animating a lump of points is harder than polygons attached to a skeleton. I don't know how they get away with reflections and lighting without some ray tracing involved. This looks really grainy and the lighting is very simplistic. The files for these games would be huge.

Even if it was posible, this kind of thing would be used by the movie industry long before it got remotely close to being able to run on a normal PC, and certainly not in realtime.

Movies use ray tracing which has superior reflections and lighting IMO. Ray tracing is still better for pre-rendered.

[ein]

The only difference between these and old 2d games is that these pixels are 3d.
And fuckin' hardcore.

But yeah, the files for these games are likely going to ungodly huge.
Like uncompressed audio.

[Armok]

It is only plausible-sounding if you don't know what they are talking abaut. And this certanly would have a lot of applications in moviemaking.

[ein]

It is only implausible-sounding if you don't know what they are talking about. And this certainly would have a lot of applications in movie-making.

Fixed that for ya.

[Armok]

Thanks!

[qwertyuiopas]

Going to their website and watching their videos that aren't compressed to youtube standards. There isn't any lag in their tech demos. Their only real issue currently seems to be getting shadows right.

Shadows are simplistic, no hint of reflection, refraction, multiple lighting, light sources(Only seems to have a single-direction sun)...

I don't know how they get away with reflections and lighting without some ray tracing involved.

All of those reflections were a single horizontal plane, and thus they can easly cheat by duplicating the scene along the plane. In fact, I think they mentioned it when talking about how many points it was calculating for.


Bottom line:
Probably plenty of FAKE.

Could probably be beat by a combination of a polygon-based system for finding the polygons, and then having it recurse into a subsystem that procedurally generates fine details like a curve of a tree, the shagginess of the trunk, a surface texture(like a current bump map, only more detailed, and generated as needed by vague parameters. Trades precision(exactly what the artist made) for detail.

Also, seen 3D modeling programs? Seen how you can create a camera flythrough in a no-textured, low-detail, technical view, hit "render", and come back a while(minutes, hours, days) later with a "real time" version in full detail?

What I am inferring is that, until a downloadable program comes out that allows the user to make their own levels, what is the diffrence between a falsified framerate and a realtime recording after it has been uploaded to youtube?

Oh, and one more thing: Size.

Did they ever show a vast landscape, stretching beyond view over the horizon?

I suspect that they can't while retaining the accuracy and details of their system, if a true system it is.



Now, despite all that, I would be delighted to be proven wrong with a mindblowing interactive downloadable example, but until/unless that happens, I will remain pessimistic.

[Gigalith]

I agree with Armok. I believe this is possible about as much as I believe perpetual motion machines are possible.

Unlimited Detail is a software algorithm that gives unlimited geometry. When we say “unlimited geometry” we really do mean it. It really is Unlimited, Infinite, endless power, for 3D graphics.

If they really had infinitely powerful graphics, then I could draw a functional CPU inside their 3D engine that's twice the size of the actual physical CPU running the machine. Since these are real-time, the inside CPU would be twice as fast as the real CPU simulating it. I would draw a hard-drive twice the size of my real hard-drive (remember, INFINITE POWER) and install two copies of the 3D engine inside of itself. Then I would write a script to do this all over again, inside that hard-drive.

Presto, infinite processing and memory. It'd be trivial to write a version of Maxwell's Daemon to break the second law of thermodynamics.

[Alexhans]

The idea certainly seems interesting.  I, as many others, have a lot of doubts about their implementation of various aspects of their system.

it's funny how they focus on the "unlimited" aspect.

Here are some other opinions:
http://www.gamedev.net/community/forums/topic.asp?topic_id=557528

[qwertyuiopas]

“unlimited geometry” Does not mean simulating it all. In fact, even if that was all realtime, did anything ever move except the camera? Couldn't they have put gigabytes of precomputation and indexing, and massive optimizing? So, “unlimited* geometry” with * being "As long as you don't mind devoting the entire CPU to drawing the static, unchanging geometry, even on the fastest computer available commercially in ten years"

Also, they claim it's all the detail the human eye can percieve. They discount that some game, somewhere will allow you to zoom in, or have other magnification devices that work ANYWHERE, and suddenly the calculation load to achieve 10x detail is just too much... (With that many points, adding more would mean more ram, and that means more cache misses, and that means massively slower, and that means that the fancy physics sim has to be cut, when a polygon based approach would still look *good enough* and leave the entire CPU free for the AI...

Now, they could instead have it procedurally generate recursive detail where the camera can see it, and yet... If they did, on a polygon-based model, you would have similar visual results, as the polygon limitation(flat, angular surfaces) is eliminated by the action of the computer generating another detail layer.


Now, how they could have implemented such technology...
Why not a giant recursive Octree, with each recursion storing the average density, and grouping rays so that you are really doing raycasting with vague mathematical definitions that later split into rays rather than individual rays?

The bonus being that you would be tracing the reflection of a geometric primitive such as a quad rather than a few thousand indevidual rays. A mirror would be a single set of grouped reflection for each geometric block that hits it, and transparency layers could be added as encountered...

But wait, there is more!

What is stopping developers from using a simple technique like this to procedurally round out low detail areas as the player approaches? "Like all hypertextures, it's really a density cloud that's been "sharpened" to look like a solid object."

[Gigalith]

So, “unlimited* geometry” with * being "As long as you don't mind devoting the entire CPU to drawing the static, unchanging geometry, even on the fastest computer available commercially in ten years"

The Unlimited Detail tree will run on anything from a PC to a mobile phone and no special graphics cards are needed.

How does it work?

If you have a background in the industry you know the above pictures are impossible. A computer can’t have unlimited power and it can’t process unlimited point cloud data because every time you process a point it must take up some processor time. But I assure you, it's real and it all works.

(emphasis mine)

I'm taking the laws of computation over mere assurations, thank you.

[Ampersand]

Like I said, I reserve judgment until the SDK is released for everyone to see for themselves. I know who the videos are intended to target; Investors. Not us.

[Eagleon]

The video disgusted me. Pretentious irritating snivelly voice. I know what a polygon is, and so do the people they're marketing this to - gamers and digital artists. Why do they feel the need to try to insult my intelligence? It's pretty obvious they could be using instanced geometry at least in part, which is a relatively mature technology. You can create scenes like the stacked pyramids using current tech, and the models in the garden scene are repeated heavily if you look carefully. Still, the concept I believe is sound - we can already identify what meshes are visible from the player's perspective, and from there it's a matter of ray-tracing per-pixel to determine what point to use. That's not that unrealistic, really, with today's processors.

And yeah, the "unlimited" stuff is marketing bullshit, and not very good bullshit. Obviously if you had very large resolutions (which are going to be more common, you really can improve things there still), the power needed is going to increase at a square or more depending on what information is needed, and computing optical properties, shadows, etc. is still a problem. I also wonder how easy it will actually be to make animated models for any of this stuff, as well as how much physical space they'll take up. I really do not relish installing a 50+ GB game, regardless of how shiny.