Break Through in Gaming Rendering Technology?

[DilTech]

Ya know... Carmack was correct.

Carmack said you can't do this on current generation hardware and that it will be years before we do. He also pointed out consoles, and it DEFINITELY won't run on current gen consoles.

http://www.youtube.com/watch?v=tkn6ubbp1SE

That's a very low resolution voxel model, ran through GPGPU, and animating it leads to only 36 fps. One model at ultra low resolution that isn't even that detailed. This was done on a GTX 460 1gb btw.

That said, notice that all the "unlimited detail" models are identical? Facing even the same direction all the way down the line? Yep... That's because they're using the same thing we've seen for years in that regard, sparse voxel octrees. Sure, they've added new techniques to the mix--but, it's still what we've already seen in that regard. Every tree is the same, and even faces the same direction.

They point out that they haven't touched the graphics card, and of course they haven't.... The GPU doesn't have support for this, and would be stuck in gpgpu mode (hence, software). The gpu barely pulled off an ultra-low resolution model walking with nothing else going on.

We probably won't see this actually put to use until we see idtech6 (probably even years past that without proper hardware support), the engine AFTER rage... which Carmack stated he found a method for unlimited polygons/geometry for (when he first introduced idtech5 he mentioned it and was surprised that thus far no one else has figured it out). In other words, without a complete market shift, then Carmack is 100% correct.

[RPGWiZaRD]

I was just thinking, software rendering has come a long way since I last was using it. :P

At first I was a bit skeptical but after watching this nice video posted by fellix_bg I'm starting to change my mind. You have to take into account this what we're seeing is far from finished but it ALREADY runs that well on a CPU alone. Also this company has existed what, maybe only 2 years and has 9 people working on it, that tells quite a lot already. I think also their way of doing to be able to scan literally reallife objects is a nice thing, imagine an elder's scrolls game with this engine. Such a variety of vegetation could be used without any performance/resource limits, it's just a matter of how much patience is put into making different objects. :p

I'm definitely looking forward to how this turns out but they got a lot to prove when it comes to animations and such still, a game isn't just static objects, leaves will have to move with the wind, ripples in water etc those kind of things are crucial for a game to feel real so that's where I'm especially interested how and if they'll be able to handle physics and such but yea basic animations is still very important too. Oh and don't forget about all those tricky lightning effects.

I'm also worried from NVIDIA and AMD's point of view how well they'll be able to utilize GPUs, because these two companies definitely aren't gonna be friends with Nucledeon otherwise and without their support, I don't see this engine taking off in a big scale.

[fellix_bg]

I hope they show us much more progress, probably after another year of absence from the public. The problem is not the memory occupancy, at least not the RAM, as their "search" algorithm picks only that much points to fill the screen. It is very easy, from performance POV, to use tons of instancing and completely static structure to show smooth fly-by real-time overview of truly unlimited geometric detail with gigabytes of point data streaming out from the HDD. Now, this provides very good data locality indeed, and traversing a static structure is almost like streaming a huge video file and seeking back and forth seamlessly. But the moment they put in some degree of interactive dynamics - lighting, shading, skinning, transforming or physics, the whole nice static structure breaks down and the search algorithm has no option, but to trash the whole memory just to find the relevant node, containing the needed data -- for every object and every frame! A robust caching can alleviate lots of the pain, but there can be that much on-chip cache to fit the all the relevant data.

[HelixPC]

http://hardocp.com/article/2011/08/1...dell_interview

Euclideon & Unlimited Detail - Bruce Dell Interview

[jogshy]

You can get the same detail using GL4.1/DX11's tessellation at the cost of 1/100 the disk or memory space you use for the voxels.

On the other hand, they only show instanced models, not many animated and deformed/morphed/skinned characters because their ray tracing structure should be then updated on the fly taking a lot of time.

I've the impression is one of those companies which simply wants to be adquired by other for some money.

[XSAlliN]

http://hardocp.com/article/2011/08/1...dell_interview

Euclideon & Unlimited Detail - Bruce Dell Interview

nice.

[gumballguy]

this company has existed what, maybe only 2 years and has 9 people working on it

Nope. See previous pic in this thread - photo of the same company in 2003! Plus whatever development time they already had. So a decade should be closer to the truth (assumedly not 9 people the whole time though)

I'm also worried from NVIDIA and AMD's point of view how well they'll be able to utilize GPUs, because these two companies definitely aren't gonna be friends with Nucledeon otherwise and without their support, I don't see this engine taking off in a big scale.

Sure, they'll stall it. But some other vendor will enter the market due to the money to be made.

[netkas]

But can they render transparent objects ?

[Kallenator]

Isn't the water transparent in the demo?

[cx-ray]

The whole interview with John Carmack is well worth watching. The below section though is specifically about Euclideon:

http://www.youtube.com/watch?v=hapCu...ilpage#t=1040s

[netkas]

>Isn't the water transparent in the demo?

I mean things like smoke, fire, windows.

it's the case when they need to find one atom (transparent one) and the atom behind it too