An Australian company claims to have developed technology to make computer game graphics "100,000 times better" than current-gen systems. Euclideon says its "Unlimited Detail" engine offers infinite geometry "makes everything out of tiny little atoms instead of flat panels". 'Atoms', it would seem is Euclideon-speak for these …
I get the feeling this is supposed to be a new, clever algorithm that can run on existing hardware.
And in at least one sense, "unlimited detail" is possible. I remember playing a game from Lucasfilm on my Commodore 64 that converted from a coarse polygon map for a given object to an equivalent finer one as one got closer to it; they billed this as "fractal graphics".
And storage for the ultimate detail of the world was avoided, since much of it was algorithmically generated at random.
So tell me this...
Aside from "production problems", just how much storage would you need to hold the maps from a typical modern game at a resolution of "64 'atoms' per cubic millimetre"??? Assuming each and every point needs, at bare minimum an RGB colour value and a light level, methinks (without so much as a beermat calculation) that the maps for an average game would probably need more storage than most sizeable research establishments have on hand, as evidenced by their small 'island' demo having 21 TRILLION polygons/data points/whatever.
It's a lovely idea in terms of prettiness but it's totally, totally unrealistic until we have storage devices with many, many thousands of times today's capacities... You don't have to look particularly closely at the example screens/vid to see that, while impressive looking, there's a hell of a lot of object repetition, which is presumably a careful way of skirting around the storage requirements problem. Everything is just a few stock objects rotated which obviously just wouldn't fly in a AAA title.
I heard the word procedural mentioned
Which generally means you don't bother creating a 1:1 "map" of the environment but instead describe sections and let the graphics engine create the ity bity details on the fly.
This probably also explains the macro level similarities and minecraft like appearance. I wouldn't read too much into that, as the guy said, they're programmers not artists and the purpose of the demo was to illustrate the levels of microscopic detail. I'd bet money that every one of the very similar looking macro blocks was in fact unique.
Procedural generation would reduce some of the footprint but still the question is how much storage did that scene require with that level of detail? Multiply it by whatever factor you'd need to increase the complexity of the scene by to make it "good enough" for a modern title and... well I still think it's going to be 'rather a lot'...
And anyway, if using procedural generation does circumvent obnoxiously large storage requirements, surely it just shifts it over onto the GPU/CPU load instead...
If Carmack isn't rubbishing it
If Carmack isn't rubbishing it after looking at it (in some depth I guess) then I don't think any of us are qualified to comment on it based on a single reg article and some light research.
After all, when it comes to Engines, Carmack knows what he is on about
nice, but there are questions
The image you picked for the article might be somewhat misleading. The beauty is in the close-ups - rocks, tree, ground. The demo "island" doesn't look very good at a gross scale.
On the other hand, the image you picked for the article in some way poses an interesting question... why does the demo island look so bad at a gross scale? I get that it takes a lot of resources to craft a good-looking square kilometer of ground, but come on, surely they could at least avoid building it out of large square blocks? This way, it makes one suspect that the blocky structure is required for the rendering engine to work efficiently, which would be a major, major problem.
Also, I'd like to know what kind of hardware it's running on. For what I know, it could be running on something so powerful that it could do exactly the same things on a standard engine.
nice when it works
In a different sphere of computing but already use point cloud data for visualisation and data editing and I can assure you it eats processing power for breakfast. It'll be great when it works, but just can't see it running in any meaningful way for a while yet. We have high spec rigs which occassionally struggle with what we'd like to throw at them, whereas games need to be capable of running on something approaching a normal home setup...
...but as with everything, a journey begins with the first few steps!
Reminds me of voxels, they were going to revolutionise games as well - their landscaping tech shat all over polygons, but because the new 3D accelerators couldn't accelerate them (not being polygons), a promising tech fell by the wayside. Comanche 4<?> and Outcast used them to great effect.
http://www.google.co.uk/search?q=outcast+game&hl=en&safe=off&rlz=1G1GGLQ_ENUK364&prmd=ivns&tbm=isch&tbo=u&source=univ&sa=X&ei=J-c4TpqMKsWj8QPFt9HgAg&ved=0CDYQsAQ&biw=1120&bih=531 (google image search for Outcast).
Voxel rendering had other issues
Namely that it was heavily patent encumbered in the US. There was no incentive to develop good voxel rendering engines or hardware when plane (badum, tish) old polygon rendering systems could be made royalty free and are perhaps simpler to envisage and implement.
Question from someone who is merely an *amateur* geek.
"id Software's John Carmack reckons there's no chance Euclideon will run on current-gen systems, but has the potential to "several years from now"."
What could this run on and why are we talking "several years"?
Other than that those graphics are gorgeous.
Wonder what hardware they will recommend to run it.
I thought 3DMark used to have a test that was made using the same sort of process, a 3D elephant if I remember correctly?
By "atom" they mean "voxel".
There's nothing new here other than the level of detail. Voxels went through a cool phase in the mid 90s, when DOS games reigned supreme. They were used in things such as Shadow Warrior (Cool! Stuff that doesn't spin around to face you when you move!) and notably in Outcast, which looked gorgeous if you had a fast CPU.
And therein lies the killer - voxels weren't hardware accelerated, weren't able to be hardware accelerated and quickly faded away once Windows gaming became the norm.
For voxels to be a success again there would have to be hardware support and even if NVidia and AMD were to start today, it'd be years before cards hit the market. In 10 years time? Who knows, but the last company that tried something different (NVidia, with its NV1 chip) ended up abandoning it in favour of the status quo.
Couldn't you, say, accelerate the voxels using (for instance) OpenCL running on the GPGPU, and then use the output of that, and still have the benefit of a massive number of very simple processing cores on the GPU?
Voxel hardware rendering support already exists
Some other guys in my company work on medical imaging systems that use off-the-shelf nvidia tesla gear attached to windows boxes for exactly that purpose. There's no nice equivalent of opengl/directx of course. As I said above (assuming it gets moderated before this post!) voxel rendering has, or had patent issues.
Still pretty cool.
Ok, as the narrator admits that they are taking a process in use by the medical community (as Ru points out) and applying it towards games, its still pretty cool.
There are other applications for 3D rendering outside of gaming, but gaming is where the money is.
It does sound like NVidia would be a good partner assuming that you could use CUDA to speed things up a bit. As Ru points out, people are doing this.
With respect to patent issues... how long does a patent last? (14 years in the US)
A lot of the basic voxel patents went out of date before 2005. Granted there are some patents that are still being granted that are based on the voxels... a quick patent search shows one :
was granted in 2007 to Intel Corporation.
It would be in Intel's best interest to actually make voxel based tech more reasonable in terms of licensing.
So definitely a thumbs Up.
Note: It could be because of patent issues that the company went dark for a while as they may have worked around some patent issues. Maybe El Reg can dig in to this?
Not quite as good as they are selling it, evidently.
Minecraft creator Marcus "Notch" Persson accuses the Euclideon team of being "snake oil salesmen,":
Well, he should know all about selling nothing.
All been done before...
Yawn. By "atom" they mean "voxel".
Nothing new here, except the level of detail. Voxels were popular a while back (there was a notable game, Outcast, which used them) but they fell out of fashion partly as they couldn't be hardware accellerated.
"A square metre of game space"
"The application of this technique with point-cloud data can pack up to 15 million converted polygons in each square metre of game space."
Well this isnt really hard given that the area is purely virtual. I can pack 15 million polygons into a square *millimetre* in my engine :) :)
I wonder how real it is.
Of course if it's really real and it really will run at a decent rate on a normal computer/console then that's all well and good but I do have to wonder how the lighting works and how dynamic it will be. How will it deal with reflections? How will it deal with movement? Will leaves rustle? Will water ripple? Will there be footprints in the dirt? These days ripples and footprints as far as I know are still done with sprites, now if they made this point-cloud thingy deform, that'd be something worth shouting about.
Pint because why isn't it Friday yet?
So.. Voxels again ?
So the fact that each voxel has XYZ and RGB data, makes it possible to fit all this data in normal memory.. how ?
(and you'll lose all that nice shader action we've been seeing more and more of in the last few years..)
Transforming these data sets tends to be a hassle
(deforming/animating polygons is easy, just move the corners of the polygon, but here the whole data set changes, where each one of those volumetric pixels. oops atoms, has to change a bit, depending on location and angle relative to the movement.)
Maybe we'll see some hybrid form pop-up again (remember the game Comanche ?) but I don't see this taking over any time soon. (shading, shadowing and collision detection between polys and voxels is a nightmare as)
I applaud every new development in graphics, but this does not look very new.
Hardware specs plz
Would like to know the hardware specs used for this demo.. this would give an idea whether it is really feasible in future..
Wondered where these guys had gone.
Not to be too pessimistic, but there are a few points:
* The storage requirements of so much detail is large, very large. Even given a very efficient sorting and filtering algorithm, you can't get over the fact that such detail requires a lot of storage.
* Animation... never seen any with this system. Without animation it's all dull and lifeless. This isn't just animation of things sliding in various planes, it's about twists and bends as well.
* Reflections - still none. The "water" effect in the video is an inverted set of different coloured objects to give the appearance of a reflection. It's an acceptable trick, but now try combining this with trick with animation of the water surface... it's meant to be water, not a mirror.
* Still no sign of transparency, another important visual effect in graphics
* The scenes are extremely repetitive, probably due to having to keep the storage requirements even partly sane. Not even the angle of placement of the repetitive objects is varied, which does seem to indicate possible restrictions of the algorithms in use.
* Still no lighting and shadows, although the end of the video did promise that there are some. In which case, why not properly demonstrate this in the video, even if only a short clip at the end?
Not to say that the tech is without promise, but it's not far from ready for replacement of current tech.
Giz a clue.
What is that running on?
Possibly a hoax
I was interested until two things caught my attention... Dell mentioend nothing about how increasing the amount of detail on the screen led to an unlimited amount of detail being available. and secondly and perhaps more telling, their website has ceased to exist.
Blame that on the register...
... Any un-prepared website the reg points to has a habit of going afk. A short while later the website was pointed to a facebook page, along with an explanation of where it went.
Well, we've had Slashdotted when Slashdot links an unsuspecting site and Googleblatted (before Google sorted that out) when Google linked a site on their search page, hmm, what can we call it when The Register does it?
I know... DeRegistered? DeRegged? Regged?
Ok, that one was bad even for me. I'm going now... >grabs coat and slinks shamefacedly out door<
Water shouldn't be square
Has strangely square shaped corners on their canals. If it were really able to do atom level accuracy, I'd expect my rivers and lakes to have round corners.
Snake oil, obv.
Notch sums it up: http://notch.tumblr.com/
Voxel-type things will find a use in hybrid approaches, i.e. for buildings and other static things that you want to deform.
Ok the proof is in the pudding and we will see what comes up in the coming year but it certainly looks impressive.
Dont think they did themselves any favours by saying they're techies not artists. Hire one in. If you're dealing with games gfx you'd think that was a given :)
I think I first heard it mentioned in the first commanche game. Good for from a distance but a little blocky up close, so they've made the blocks smaller now? Is that it?
sounds a lot like
Voxels to me.
So, Voxels, then
This is basically what this is, surely?
Invisible magic computer!
I would like to see the kind of gear that runs this in anything approaching real time. Working through 20 tera-polygons is no mean feat.
Realistically, compare with: http://artis.imag.fr/Publications/2009/CNLE09/
looks good but:
So, highly advanced voxels then?
The 1990s called...
They want their Voxels back.
Nice technique for static scenes. However, heavy reuse of object geometry is required to prevent the scene becoming huge in storage terms. Also, character animation is tricky, so a hybrid system would be required for any game that isn't a tree or landscape simulator.
Like the Murphy's, he's not bitter.
Add physics and it will fall over...
Lets see them add some physics into this and watch even the fastests quad GPU rig fall over.
No reason to get out of bed...
because the real world will be so boring. Jack me in!
If they want to succeed...
They need to fire that narrator.
If they can do physics on the atoms it's more like a simulation than a game engine, Lloyd Grossman was a nice touch as well.
Is this really new?
If I understand the description correctly this sounds like its just tiny voxels with a tool for generating them from polygon data?
Of course I may misunderstand.
That could be the key...
That is, if they can convert polygon data to voxellated objects on the fly.
Think about that - if it's doable, then this really would revolutionise game graphics. You'd store a polygon version of a palm tree (which takes very little space), along with a procedural dataset for generating bark and leaf surfaces when you get close enough. Then your converter kicks in and renders out the voxel tree. Then you just do instancing of your stored objects, and either store or procedurally generate a map of where your instanced copies are in the game world. Storage problem solved.
This would also eliminate the previously-mentioned animation, deformation and physics issues as well - you do your deformation and physics on the stored polygon objects, then convert them over to voxel objects on the fly - and off we go!
Just in case this is NOT what Euclidion are doing, if what I've described above hasn't been done yet, I hereby declare this post as prior art in the event of any greedy corporate pigs looking to patent it down the track, and release it into the public domain as an 'open patent' (i.e. you can use it freely but you can't stop anyone else from using it or charge them for doing so.) :-)
I think we can handle the word voxels.
They could do with someone else doing the voice-over too (see the From Dust tech demo http://www.youtube.com/watch?v=CfKQCAxizrA)
All very impressive, but I'll be convinced when I can try out a demo on my graphics card.
So basically, they reinvented voxels?
Not doing themselves any favours...
The technology sounds interesting and has the potential to revolutionise graphics-intensive applications... but i can't help but be sceptical about a company that can't even maintain a functioning website.
Nonetheless, the best of luck to them!
God I hope these guys are for real!
One thing notable about the demo is there is no animation or physics. There is also a lot of repeated content cut and pasted over and over lined up in neat little x / y rows which could have a major impact on memory and transformations.
I think when we see a demo consisting of procedurally generated trees on a random landscape with some measure of animation (swaying trees, flags, a few NPCs walking around etc) then we're talking about a potentially viable solution.
At the moment it could just be so much smoke and mirrors.
But is all of that stuff just scenery or can we interact with it? Can I pick that rock up and use it as a weapon? Or pick up a handful of gravel and blind an opponent with it? I really really hope so.
- Opportunity selfie: Martian winds have given the spunky ol' rover a spring cleaning
- Spanish village called 'Kill the Jews' mulls rebranding exercise
- NASA finds first Earth-sized planet in a habitable zone around star
- New Facebook phone app allows you to stalk your mates
- Battle of the Linux clouds! Linode DOUBLES RAM to take on Digital Ocean