That's why we need decent procedural modelling tools so that artists don't need to be polygon technicians. Since I last saw this a little over a year ago, it does seem they're making some progress in terms of integration with normal artist workflow, although I'm still skeptical for various reasons. I don't know enough to fully judge, though.
I'm digging up this old thread because The folks over at [H]ardOCP got a video interview . There's a real time demo at 22:25 running on an Asus G73SW (same laptop I have ). edit: I actually watched the entire 41 minutes of it. I must admit that the horrible quality of the recording in the beginning and the fact that they were only rerunning old filmed stuff in the background made me expect Rick Astley to pop out any second, but he never did. Bruce Dell really doesn't give anything new away about how the technology works in the video. I see this mainly as a "proof that Euclideon does exist and its software is real" kind of thing. I think the most interesting question (and the one that he acted most nervous when answering) was the question about memory usage. He's still giving that vague "we compress it" answer to that question which I find hard to believe. How could they possibly compress what must be several terabytes of raw data into something that can be processed on 8GB of memory? Using his example of a library, you can index all the books and create an army of crawlers to find the one you want instantly, but you still need all the shelves to hold all the books in.
Very interesting video. I'm a bit confused as to how they create fictional objects though as everything in that demo has been scanned in. Anyone who understood care to explain?
Well he said that it may influence the artists to go to mediums such as clay to make their models and then scan them in. Very interesting video indeed, certainly proves that it isn't a rendered video.
He can't, period. The only solution for this is not to use many different types of object, but just few. For example only few types of trees, small block of grass repeated over and over,... And he is still in the volume of stuff which doesn't fit in RAM.
Nothing is impossible in this world, if you set your mind to it. As I have stated before 'If it wasn't for our imaginations, we wouldn't have the things we have today'. Think about it, if someone said to you back in the 80s "we will have handheld computers soon" (Tablets, etc) you would have laughed in their face. So now lets say time machines are impossible, what happens in few hundred/thousdand years time when someone cracks that? Does time really exist? or is it just something we gauge our self on...
There are technological limits. You need the data stored somewhere, and that data is not going to be small. I love how people defending this technology use abstract words like "progress, nothing is impossible", while the ones who understand know they can't do it unless our memory and hard drive space expand in unprecedented rate. Just think about it - to have a point somewhere in space, you need to store it's location, let's do it with integers. That means 2 bytes for X, 2 bytes for Y, 2 bytes for Z. You need to store the color of the point as well, that means another for bytes - red, green, blue, alpha. So we are at 10 bytes for just 1 point. Make the resolution of the item at 0.1mm, that means 10 points per milimeter, 10x10 = 100 bytes per milimeter. Surface area of a human skin is approx 2 square meters, that means you need to store 2000x2000 = 4 000 000 points of information, which means 40000000 bytes of data just for one person. That is 38,14MiB for just one small, static model of human. Now do it in bigger scale, let's do a surface of 1 square kilometer, with nothing on it, just a simple, plain surface. that is 1 000 000x1 000 000 milimeters = 1 000 000 000 000 x 10 = 10 000 000 000 000 bytes of data = 9,09 TiB of data - for a simple 1km2 plane surface. Show me a compression algorithm which solves this. And we didn't even started talking about movement, animation, body and general game physics... PS: Integers won't be enough for the 1km2 surface, so you need at least longs = 3x4+4x1 bytes = 16 bytes per point.
Instead of jumping up and down pointing fingers and yelling that's impossible, maybe we should just wait and see what they come up with when they decide to resurface with the finished product and base our conclusions on that. Even in that video there is a good 20 different models probably with the 2 statues, the trees the leaves the rocks the dirt and various other clutter. But at no point did I see a 1km^2 flat surface, from what I can see the ground is made up of the same model repeated, that will have certainly saved some space. If each model does use that much space I agree, there is no way that you could make a game out of the technology because that would place the demo at roughly 100GB of different models and you would want many more than 20 models in your game. As I said earlier, I'm just going to wait and see.
That is exactly the point. If you reuse your models, you can save a lot of space - but you can't reuse them on the scale like they did on the demo, you will need hundreds of models just for the flora, then a huge model for the surface itself (and 1km2 surface is not big), not even talking about moving objects... Simply, the needs for disk space are enormous for anything more than a technological demo with few models. So unless we are going to have 1000TB+ hard drives by 2016, i don't see how this could work for anything more than miniature demo worlds.
If only faugusztin had been able to put David Braben right on a few things before he wrote Elite... (sorry for the bump)
What does this have to do anything with Elite ? What does a game with line graphics have anything to do with a engine which is supposed to have 64 "atoms" per milimeter according to Euclideon? That information has to come from somewhere - and no one invented a losless compression method which would be as effective as this engine would need. Of course if they store lines between 2 points instead of real pixel perfect information for every atom in that line, they are able to save large chunks of memory - but then stop talking about unlimited detail...
Well, I mention Elite of course because of procedural generation and as a flippant example of something with lots of information packed into a tiny space, not because of the rendering method. You could make precisely analogous arguments about how it would never be possible to put such a thing on a floppy disk... How much information does the mandelbrot set contain? How much detail? Of course, since they are talking modelling with scanned real-world data it's not like they're just synthesising abstract things from scratch so they must be doing something at least close to general compression, but they could still have some kind of high-level parameterisation that is able to produce an overrall shape approximating a given scanned input while also providing lots of - more arbitrary - detail and not requiring that much data to specify it. In other words, who says it needs to be lossless? It can be highly lossy and still provide 'unlimited detail'. Not sure why I felt the need to make it personal, I hope you can look past that... the point is I think it's genuinely wrong to assume that you can start bandying about numbers as proof when we don't really know what methods they are using. edit: please correct me if they did indeed state to be using lossless compression...
They don't give away any technical info, that is why pretty much no one trusts them. And now they either use atoms like they do and then they should't do any fancy optimisations like "well, this is straight line, so we define it as a line instead of set of points" (same for curve or other object) - because then it is a little more detailed and lot less flexible polygon engine. And still no one gave me an alternative to the insane memory requirements of representing such massive amount of points. The objects they presented were neither big or complicated - yeah, there was the elephant and some trees - but for all elephants you need just one copy in memory - one copy of elephant, one copy of tree, one copy of grass. In real game, you will need hundreds of those, plus some very big ones like the floor or houses. And i can't imagine putting "unlimited detail" together with anything else than lossless compression techniques, and for high speed you can't use any high level compression... I'm really looking forward to technical explanation of the memory requirements and how they expect to create games like Half Life 2 for example in their engine, even with the detail level of original HL2.
