A couple weeks ago a video made the rounds of the internet and most people called this bullshit and impossible.
This is a recent 1hr follow up interview from the snake oil salesman himself. I would be keen to hear any feedback to this.
Euclideon "Unlimited Detail " engine
Moderator: Coders of Rage
- Milch
- Chaos Rift Junior
- Posts: 241
- Joined: Sat Jul 11, 2009 5:55 am
- Programming Language of Choice: C++
- Location: Austria, Vienna
Re: Euclideon "Unlimited Detail " engine
I'm not sure about this, it would be cool if it was real, but it just seems fake, because he simply keeps on talking on how great it is and whatnot and I cant imagine that there are no drawbacks at all.
But I think its cool that they show new footage
But I think its cool that they show new footage
Follow me on twitter!
-
- Respected Programmer
- Posts: 387
- Joined: Fri Dec 19, 2008 3:33 pm
- Location: Dallas
- Contact:
Re: Euclideon "Unlimited Detail " engine
Alright, a few words on this. I remember talking about this in IRC.
This sort of technology does exist, but not to the scale that he's claiming that it exists. It is true that you can render virtually unlimited LOD voxel environments that are constantly random. In other words the technology does not exist yet to render discrete worlds to an unlimited level of detail simply because the interpolation between cloud/point data and the searching algorithms are just not fast enough to be done on current hardware.....ESPECIALLY not a CPU.
Modern implementations that HAVE approached this problem actually do leverage the GPU heavily because it's already optimized to perform the operations particular to this problem very very quickly. It's a multicomputer so it clearly makes sense to make use of that kind of processing power. I simply don't see an i7 doing this. But essentially what is happening is that you have a particular set of data, which is going to be a significant amount of data. To garner a respectable resolution such that the interpolation algos dont fall on their faces you'd need several gigs of data (their implementation). But the goal is to interpolate the data down to where you have a polygon that is either a pixel or less than a pixel in size. So the voxels are actually discrete geometry because thats all we can do on modern hardware, but this is fine and they can be shaded in real time no problem (we already do this).The doubt comes in when you begin to search through the point/cloud data for accessible geometry and when you begin to interpolate at the pixel and subpixel levels. The most respectable graphics hardware still can't do this yet at any respectable speed. The most promising techniques, such as Carmack's SVO, are a hybrid approach. Here you can approach unlimited LOD to a point.
Is this where we're going? It sort of looks like it. Future hardware will probably take a hybrid approach to rendering. There's no reason to throw out whats working well for us now which is why Carmack's approach looks like it could be a winner because ultimately he relies on tessellation. The approach in the video is very 1977 Computer Graphics text-ey :P If such a word exists. They're performing their "magic" in a VERY classical sense which we know now is simply far too brute force to work on current hardware.
The playable demo is VERY clear to me that this is NOT a voxel terrain. Whatever he is showing here looks clearly like a completely tessellated world with infinite LOD texturing. Which really isn't infinite at all, but appears to be close enough. This technique HAS been implemented by many and is currently employed by several technologies. This technique is megatexturing and was pioneered by Carmack and subsequently quickly duplicated.
I would say that the guy is full of shit, but only partly. The technology exists, but not quite to that scale yet.
This sort of technology does exist, but not to the scale that he's claiming that it exists. It is true that you can render virtually unlimited LOD voxel environments that are constantly random. In other words the technology does not exist yet to render discrete worlds to an unlimited level of detail simply because the interpolation between cloud/point data and the searching algorithms are just not fast enough to be done on current hardware.....ESPECIALLY not a CPU.
Modern implementations that HAVE approached this problem actually do leverage the GPU heavily because it's already optimized to perform the operations particular to this problem very very quickly. It's a multicomputer so it clearly makes sense to make use of that kind of processing power. I simply don't see an i7 doing this. But essentially what is happening is that you have a particular set of data, which is going to be a significant amount of data. To garner a respectable resolution such that the interpolation algos dont fall on their faces you'd need several gigs of data (their implementation). But the goal is to interpolate the data down to where you have a polygon that is either a pixel or less than a pixel in size. So the voxels are actually discrete geometry because thats all we can do on modern hardware, but this is fine and they can be shaded in real time no problem (we already do this).The doubt comes in when you begin to search through the point/cloud data for accessible geometry and when you begin to interpolate at the pixel and subpixel levels. The most respectable graphics hardware still can't do this yet at any respectable speed. The most promising techniques, such as Carmack's SVO, are a hybrid approach. Here you can approach unlimited LOD to a point.
Is this where we're going? It sort of looks like it. Future hardware will probably take a hybrid approach to rendering. There's no reason to throw out whats working well for us now which is why Carmack's approach looks like it could be a winner because ultimately he relies on tessellation. The approach in the video is very 1977 Computer Graphics text-ey :P If such a word exists. They're performing their "magic" in a VERY classical sense which we know now is simply far too brute force to work on current hardware.
The playable demo is VERY clear to me that this is NOT a voxel terrain. Whatever he is showing here looks clearly like a completely tessellated world with infinite LOD texturing. Which really isn't infinite at all, but appears to be close enough. This technique HAS been implemented by many and is currently employed by several technologies. This technique is megatexturing and was pioneered by Carmack and subsequently quickly duplicated.
I would say that the guy is full of shit, but only partly. The technology exists, but not quite to that scale yet.
- Ginto8
- ES Beta Backer
- Posts: 1064
- Joined: Tue Jan 06, 2009 4:12 pm
- Programming Language of Choice: C/C++, Java
Re: Euclideon "Unlimited Detail " engine
notch had some things to say about this: http://notch.tumblr.com/post/8386977075/its-a-scam and http://notch.tumblr.com/post/8423008802 ... not-a-scam
Quit procrastinating and make something awesome.
Ducky wrote:Give a man some wood, he'll be warm for the night. Put him on fire and he'll be warm for the rest of his life.
- THe Floating Brain
- Chaos Rift Junior
- Posts: 284
- Joined: Tue Dec 28, 2010 7:22 pm
- Current Project: RTS possible Third Person shooter engine.
- Favorite Gaming Platforms: PC, Wii, Xbox 360, GAME CUBE!!!!!!!!!!!!!!!!!!!!!!
- Programming Language of Choice: C/C++, Python 3, C#
- Location: U.S
Re: Euclideon "Unlimited Detail " engine
Personally I still question how they even load their scanned model in that fast and transform it into "atoms" so fast. But there are a lot of things in history that people said were impossible that happened. So I figure we just have to wait and see.
"Why did we say we were going to say we were going to change the world tomorrow yesterday? Maybe you can." - Myself
- Ginto8
- ES Beta Backer
- Posts: 1064
- Joined: Tue Jan 06, 2009 4:12 pm
- Programming Language of Choice: C/C++, Java
Re: Euclideon "Unlimited Detail " engine
I watched about half the interview, and then I got fed up. It sounds like an infomercial! He takes what Notch and Carmack say out of context to cancel each other out. He's got a silver tongue, that's for sure, but that's all I can be sure about. I think if he actually released this "fantastic algorithm", he'd be given a lot of credit, and, rather than what he says, he'd have his approach improved upon, not broken down as something sub-par! It's his jealous secrecy that is getting him in trouble! That kind of secrecy has happened before in the computer industry, many times, and it has always bottlenecked the progress of the areas the secrets are in! If he took this algorithm, released it, then made a super-awesome-fantastic game from it, he'd be great! But no, he protects the algorithm, trying to use it as his sole source of riches and glory, when, if he released it, he'd have all the glory, and plenty of opportunity to have the riches!
And all that on the CPU? You can't run 2D software-rendered games at a decent framerate!
</rant> he talks like a politician...
And all that on the CPU? You can't run 2D software-rendered games at a decent framerate!
</rant> he talks like a politician...
Quit procrastinating and make something awesome.
Ducky wrote:Give a man some wood, he'll be warm for the night. Put him on fire and he'll be warm for the rest of his life.
-
- Respected Programmer
- Posts: 387
- Joined: Fri Dec 19, 2008 3:33 pm
- Location: Dallas
- Contact:
Re: Euclideon "Unlimited Detail " engine
Yes, you absolutely can.And all that on the CPU? You can't run 2D software-rendered games at a decent framerate!
- EccentricDuck
- Chaos Rift Junior
- Posts: 305
- Joined: Sun Feb 21, 2010 11:18 pm
- Current Project: Isometric "2.5D" Airship Game
- Favorite Gaming Platforms: PS2, SNES, GBA, PC
- Programming Language of Choice: C#, Python, JScript
- Location: Edmonton, Alberta
Re: Euclideon "Unlimited Detail " engine
I'm pretty sure most emulators don't take advantage of hardware acceleration at all. On top of that, there's an extra layer of abstraction for the different instruction sets on other hardware. I can still get 2D games and 3D games like Ocarina of Time running smoothly.