Candescence
Member
Ever since the most recent news, I've been thinking this thread might be necessary, to, firstly, dispel some of the hyperbole and misconceptions surrounding this whole thing, to straighten up the facts somewhat, and secondly, collate news and information about Unlimited Detail into one place. If some mod feels that's unnecessary, feel free to lock, I suppose. I'm gonna keep most of the videos and news updates to the bottom of the OP, unless I need certain links to demonstrate certain things. I hope this thread is at least somewhat informative.
Euclideon is a graphics middleware company based out of Brisbane, Australia, founded by Bruce Dell, specializing in point cloud/voxel data. And also infamous for overhyping the capabilities of their main project, "Unlimited Detail", right down to the name. They are terrible at properly selling the idea, unfortunately, because, provided that it works properly and is actually game-ready, is potentially groundbreaking. So I'll try and 'sell' it for them, in a much less hyperbolic manner.
Disclaimer: Not a graphics expert or engineer or anything of the sort, just a university student studying games design, so my explanation may possibly be inaccurate on the technical side. You're better off asking someone else for a more technical explanation if you want one. I'm pretty sure all this is about right, but this is just to be on the safe side.
It is, according to its developers, a highly sophisticated voxel/point cloud rendering system that can render voxel environments far, far more efficiently than polygon engines.
Here's the basics of what Unlimited Detail is in visual form:
Compared with a relatively recent game screenshot:
What you're seeing in the lower image is slightly outdated, but the improvements to environment geometry aren't really substantial in more recent titles. Take note of the ground - nearly everything is flat, with the occasional small detail object, and the grass are literally billboards, flat textured objects designed to create the illusion of grass without adding much to the polygon count. Even the player model and weapon has some polygon limits if you know where to look, though much less obvious. By contrast, the upper screenshots have thousands upon thousands of highly detailed objects, from the grass, to the individual bits of dirt in the ground. Composition and colour palette aside, the actual detail far obviously exceeds the lower screenshot. This exact demo has been demonstrated on-camera in playable form.
How does it work? "Unlimited Detail", as far as we know (putting aside the speculation of the exact details of how the voxels are structured), utilizes what Bruce Dell calls an efficient 'search engine' that, with every frame, acquires every visible voxel on-screen that can be assigned to a pixel on the screen, ignores the rest, and constructs the image accordingly. It is also claimed and shown that this technology can be run at 30FPS in software on a pre-2012 non-gaming laptop. This is incredibly efficient, and, provided that it works as previously stated, capable of incredible resolutions and an extreme amount of detail with very little processing required, relatively speaking by today's standards. It is also stated and shown that both UD voxels and polygons can coexist in the same scene.
Depending on your view, the name is slightly disingenuous - while there are still theoretical hardware and storage limits (mainly RAM and storage data), and the amount displayed is limited by the monitor resolution, the amount of voxel data in a scene at one time has literally no bearing on how efficient the actual renderer is, in theory, the renderer will always collect and display only the voxels that can possibly be visible on-screen each frame, no more, no less. Once again, we lack the actual knowledge of the limits of the tech, but according to interviews with geospacial data companies who have actually used the Geoverse offshoot, they have tried and failed to make the engine fail or even falter.
Speaking of which, Euclideon released a variant of the technology, called Geoverse, as a data viewer for geospacial LiDAR data - in other words, entire environments scanned in 3D.
Also speaking of which, one of the main features of the technology is the ability for developers to scan in objects from the real world (quality is dependent on the scanning tech, keep in mind, a focused scan on a single object is going to look a lot better than a total environment scan) without having to degrade the end result to fit a traditional game engine. It has also been stated that there are pipelines for traditional modelling workflows, enabling the use of extremely high-quality models such as the stuff you see made in ZBrush. Scenes can be constructed like any regular game world, and entire environments can be scanned in from the real world, albeit in debatable quality due to the limits of LiDAR scanning.
In conclusion: Is the technology that amazingly revolutionary? Depending on how you view it, no, it really isn't, far from it, voxel tech has been around for a long time in various forms. However, much like what Oculus is doing for VR, "Unlimited Detail" could very well potentially make pure voxel rendering practical for high-detail game worlds. If it works well enough, it could also potentially render polygon-based engines effectively obsolete. Potentially. The jury is still very much out on that.
Fun fact: Bruce Dell began developing the technology when he accidentally mistook Donkey Kong Country's pre-rendered CGI graphics for real-time visuals - basically, his reaction was "challenge accepted". He even started on his trusty Amiga. It was, to my understanding, mainly a hobby project until he actually was able to form a company and start developing it more seriously.
Aside from what was noted above, there are other important aspects that are still missing from the picture. The first three of these have been previously noted as the most important factors by Euclideon in terms of making the technology 'game-ready'. (If I've missed something, let me know.)
Until we know that these factors have been adequately addressed, we don't know if the technology is ready for gaming. However, in theory, what we know so far can still be used with static environments mixed with polygon objects. But, once again, it's "wait and see" at this point.
Q: Isn't this a scam/hoax?
A: At this point? No, not really. Euclideon have put out an actual product (Geoverse) that clients are using, and responding positively to. Now, is the tech actually 'game-ready' and if they'll release such a game-ready middleware at all? The jury is still out on that one, but Euclideon are reportedly developing two game projects to demonstrate the technology.
Q: But Notch said it was! And John Carmack said it's way too early for such tech to work!
A: Even experts can be wrong (and it can be argued that Notch isn't really an 'expert'). Considering the hyperbolic sales pitch and the 'too good to be true' saying, I don't really blame them, though. I'm not going to get into specifics here, however, and it's too early to call a verdict.
Q: But Euclideon are lying and over-exaggerating how awesome "Unlimited Detail" is!
A: Ignore the sales pitch if you can help it, as previously said, their sales pitches for the tech are cringeworthy a lot of the time. Not so much 'lying' but rather just overhyping and hyperbole. Strip away all that and focus on the facts, and there's still a lot to like.
Q: I've seen the videos with the real-world environment scans, if it's supposedly 'unlimited', why does it look kinda awful?
A: As previously stated, it's to do with the data, not the renderer. The scans are the end result of moving scanners that are trying to collect as much data as possible in an environment, and as such, is only capable of collecting so much data. Such scanners also usually remove or mess up moving objects, and especially small details - you can see how this affects plants and other objects affected by wind. In the 2011 demo, the laser-scanned objects look much better and cleaner due to focused, controlled scanning, and there is much more precise scanning tech out there these days. Even if the data has flaws, the sheer amount of geometry still eclipses modern engines.
Q: I've heard from somewhere that it's cloud-based.
A: No, it isn't. That's just one way Geoverse can load in data from external storage.
Q: The 2011 demo looks awfully repetitive...
A: That's because it was cobbled together in a few weeks ahead of time for Gamescom. The repetitiveness is still irrelevant anyway, it's still rendering a ton of detail in the actual scene. Oh, and the 'water' is actually the entire scene mirrored and recoloured, meaning that's double the potential geometry to render.
Q: But it's still too good to be true!
A: Perhaps. While it's sensible to keep one's expectations in check, sometimes new thinking can overturn common wisdom. Keep the hype low, and wait until we have something playable to judge. And even then, it might improve over time, so who knows?
Q: You're only giving them the benefit of the doubt because you're Australian!
A: What? No. That's silly.
Q: Okay, let's assume that it does work as you've described. What kind of stuff would it enable, aside from catapulting real-time 3D graphics at least a couple of console gens?
A: Well, I suppose I can make a list... (Note: pretty much speculation on my part.)
1994: Bruce Dell starts working on the tech after mistaking Donkey Kong Country's pre-rendered CGI, using his Amiga. The result is colourful, unrealistic and a bit crude/primitive, but still shockingly detailed real-time 3D for the time.
2010: Dell founds Euclideon as a company, and releases a few videos on Youtube showing a much earlier version of the tech with no colours, including a point cloud animation test. Disappears into the woodwork for about a year.
2011:
2013:
Comment on the latest news: I have no plans on making any predictions on when and what these guys are going to pull something out next, I've tried and failed before. But I have to wonder what kind of game projects they're working on. I know Dell has previously stated that they were working on another playable demonstration with the help of ex-THQ artists, perhaps one of those projects is what they were working on. Who knows. But apparently they're fully focused on gaming content, now that they've finished Geoverse.
I'm personally cautiously optimistic, hoping for the best, bracing for the worst. I really want to see it happen and working, but I'm also worried about being disappointed. Oh, well, that's what happens when you're trying not to get too excited about a seemingly far-fetched but amazing-looking project.
Euclideon is a graphics middleware company based out of Brisbane, Australia, founded by Bruce Dell, specializing in point cloud/voxel data. And also infamous for overhyping the capabilities of their main project, "Unlimited Detail", right down to the name. They are terrible at properly selling the idea, unfortunately, because, provided that it works properly and is actually game-ready, is potentially groundbreaking. So I'll try and 'sell' it for them, in a much less hyperbolic manner.
You're welcome, you wankers.
Disclaimer: Not a graphics expert or engineer or anything of the sort, just a university student studying games design, so my explanation may possibly be inaccurate on the technical side. You're better off asking someone else for a more technical explanation if you want one. I'm pretty sure all this is about right, but this is just to be on the safe side.
It is, according to its developers, a highly sophisticated voxel/point cloud rendering system that can render voxel environments far, far more efficiently than polygon engines.
Here's the basics of what Unlimited Detail is in visual form:
Compared with a relatively recent game screenshot:
What you're seeing in the lower image is slightly outdated, but the improvements to environment geometry aren't really substantial in more recent titles. Take note of the ground - nearly everything is flat, with the occasional small detail object, and the grass are literally billboards, flat textured objects designed to create the illusion of grass without adding much to the polygon count. Even the player model and weapon has some polygon limits if you know where to look, though much less obvious. By contrast, the upper screenshots have thousands upon thousands of highly detailed objects, from the grass, to the individual bits of dirt in the ground. Composition and colour palette aside, the actual detail far obviously exceeds the lower screenshot. This exact demo has been demonstrated on-camera in playable form.
How does it work? "Unlimited Detail", as far as we know (putting aside the speculation of the exact details of how the voxels are structured), utilizes what Bruce Dell calls an efficient 'search engine' that, with every frame, acquires every visible voxel on-screen that can be assigned to a pixel on the screen, ignores the rest, and constructs the image accordingly. It is also claimed and shown that this technology can be run at 30FPS in software on a pre-2012 non-gaming laptop. This is incredibly efficient, and, provided that it works as previously stated, capable of incredible resolutions and an extreme amount of detail with very little processing required, relatively speaking by today's standards. It is also stated and shown that both UD voxels and polygons can coexist in the same scene.
Depending on your view, the name is slightly disingenuous - while there are still theoretical hardware and storage limits (mainly RAM and storage data), and the amount displayed is limited by the monitor resolution, the amount of voxel data in a scene at one time has literally no bearing on how efficient the actual renderer is, in theory, the renderer will always collect and display only the voxels that can possibly be visible on-screen each frame, no more, no less. Once again, we lack the actual knowledge of the limits of the tech, but according to interviews with geospacial data companies who have actually used the Geoverse offshoot, they have tried and failed to make the engine fail or even falter.
Speaking of which, Euclideon released a variant of the technology, called Geoverse, as a data viewer for geospacial LiDAR data - in other words, entire environments scanned in 3D.
Also speaking of which, one of the main features of the technology is the ability for developers to scan in objects from the real world (quality is dependent on the scanning tech, keep in mind, a focused scan on a single object is going to look a lot better than a total environment scan) without having to degrade the end result to fit a traditional game engine. It has also been stated that there are pipelines for traditional modelling workflows, enabling the use of extremely high-quality models such as the stuff you see made in ZBrush. Scenes can be constructed like any regular game world, and entire environments can be scanned in from the real world, albeit in debatable quality due to the limits of LiDAR scanning.
In conclusion: Is the technology that amazingly revolutionary? Depending on how you view it, no, it really isn't, far from it, voxel tech has been around for a long time in various forms. However, much like what Oculus is doing for VR, "Unlimited Detail" could very well potentially make pure voxel rendering practical for high-detail game worlds. If it works well enough, it could also potentially render polygon-based engines effectively obsolete. Potentially. The jury is still very much out on that.
Fun fact: Bruce Dell began developing the technology when he accidentally mistook Donkey Kong Country's pre-rendered CGI graphics for real-time visuals - basically, his reaction was "challenge accepted". He even started on his trusty Amiga. It was, to my understanding, mainly a hobby project until he actually was able to form a company and start developing it more seriously.
Aside from what was noted above, there are other important aspects that are still missing from the picture. The first three of these have been previously noted as the most important factors by Euclideon in terms of making the technology 'game-ready'. (If I've missed something, let me know.)
- Animation: The above-linked 2011 interview with Bruce Dell includes rudimentary animation from previous iterations of the technology. There's also an animation test video from about four years back. However, we have yet to see proper animation from the current iterations of the tech.
- Dynamic Lighting: Actually demonstrated in a mini-demo at Gamescom 2011, also demonstrating how UD voxels can coexist with polygons.
- Collisions and Physics: Unknown. However, theoretically, it could easily be achieved the 'traditional' way, though one can imagine the aim is to create a system that works better for voxel-based environments.
- Data storage and RAM: Also somewhat unknown, though Bruce Dell has stated that it is also a major focus. Considering Geoverse was also designed to greatly optimize LiDAR data, who knows what the UD format might be like.
- Water: The 2011 demo doesn't actually have water, that's just an illusion caused by the entire scene being mirrored and tinted (which is still a big deal, in other engines, that trick would be considered completely insane and wasteful). That being said, it's not difficult to conceive water rendering in UD anyway, especially if they get animation working in general.
Until we know that these factors have been adequately addressed, we don't know if the technology is ready for gaming. However, in theory, what we know so far can still be used with static environments mixed with polygon objects. But, once again, it's "wait and see" at this point.
Q: Isn't this a scam/hoax?
A: At this point? No, not really. Euclideon have put out an actual product (Geoverse) that clients are using, and responding positively to. Now, is the tech actually 'game-ready' and if they'll release such a game-ready middleware at all? The jury is still out on that one, but Euclideon are reportedly developing two game projects to demonstrate the technology.
Q: But Notch said it was! And John Carmack said it's way too early for such tech to work!
A: Even experts can be wrong (and it can be argued that Notch isn't really an 'expert'). Considering the hyperbolic sales pitch and the 'too good to be true' saying, I don't really blame them, though. I'm not going to get into specifics here, however, and it's too early to call a verdict.
Q: But Euclideon are lying and over-exaggerating how awesome "Unlimited Detail" is!
A: Ignore the sales pitch if you can help it, as previously said, their sales pitches for the tech are cringeworthy a lot of the time. Not so much 'lying' but rather just overhyping and hyperbole. Strip away all that and focus on the facts, and there's still a lot to like.
Q: I've seen the videos with the real-world environment scans, if it's supposedly 'unlimited', why does it look kinda awful?
A: As previously stated, it's to do with the data, not the renderer. The scans are the end result of moving scanners that are trying to collect as much data as possible in an environment, and as such, is only capable of collecting so much data. Such scanners also usually remove or mess up moving objects, and especially small details - you can see how this affects plants and other objects affected by wind. In the 2011 demo, the laser-scanned objects look much better and cleaner due to focused, controlled scanning, and there is much more precise scanning tech out there these days. Even if the data has flaws, the sheer amount of geometry still eclipses modern engines.
Q: I've heard from somewhere that it's cloud-based.
A: No, it isn't. That's just one way Geoverse can load in data from external storage.
Q: The 2011 demo looks awfully repetitive...
A: That's because it was cobbled together in a few weeks ahead of time for Gamescom. The repetitiveness is still irrelevant anyway, it's still rendering a ton of detail in the actual scene. Oh, and the 'water' is actually the entire scene mirrored and recoloured, meaning that's double the potential geometry to render.
Q: But it's still too good to be true!
A: Perhaps. While it's sensible to keep one's expectations in check, sometimes new thinking can overturn common wisdom. Keep the hype low, and wait until we have something playable to judge. And even then, it might improve over time, so who knows?
Q: You're only giving them the benefit of the doubt because you're Australian!
A: What? No. That's silly.
Q: Okay, let's assume that it does work as you've described. What kind of stuff would it enable, aside from catapulting real-time 3D graphics at least a couple of console gens?
A: Well, I suppose I can make a list... (Note: pretty much speculation on my part.)
- No need for flat geometry. You can probably even convert bump mapping to actual, proper geometry. And, for that matter, no more visible polygon vertexes.
- Proper grass and flora that isn't a bunch of flat textures that, depending on the game, may or may not constantly face towards the camera.
- New potential avenues of content creation, and potentially cheaper. Real-world objects can be scanned with no need to compromise on quality.
- Models don't need to be remade for more advanced hardware if they're of high enough quality. This is especially true of 'realistic' models.
- Level-of-detail (LOD) models are essentially obsolete.
- We could basically have games on the level of CGI. Actually, thinking about it, it could also enable real-time non-interactive animations, if mainly for the novelty of watching them in 60 FPS.
- Oculus Rift-compatible UD games would be much, much easier to optimize for 120 FPS.
- 4K, 8K and 3D-view games without the need for high-end PCs.
- The efficient rendering would leave a lot of room for other processing-intensive features, such as lighting, physics, and fluid dynamics.
- Destructible geometry with everything is much, much much easier, and much more dynamic.
- And other stuff I'm probably forgetting.
1994: Bruce Dell starts working on the tech after mistaking Donkey Kong Country's pre-rendered CGI, using his Amiga. The result is colourful, unrealistic and a bit crude/primitive, but still shockingly detailed real-time 3D for the time.
2010: Dell founds Euclideon as a company, and releases a few videos on Youtube showing a much earlier version of the tech with no colours, including a point cloud animation test. Disappears into the woodwork for about a year.
2011:
- Euclideon pops up again, with another demonstration that was cobbled together in a few weeks in time for Gamescom. Goes viral instantly, and also causes a lot of people to cry 'scam'.
- John Gatt from GamingFace actually decides to get an interview with Bruce Dell at Euclideon, and gets an on-camera, hands-on demonstration of the demo, proving that it isn't actually pre-rendered. Also includes some other various interesting tibits, including clips of much older iterations of the tech, which also show rudimentary animation.
- Euclideon shows up at Gamescom that year, showing off the demo as well a s dynamic lighting demo. Also notable for actually getting a lot of praise from Cevat Yerli, the head of Crytek.
- GameInformer does an in-depth interview feature on the technology, including an interview with Bruce Dell. Interesting titbits include graphics card makers interested in implementing special tech to accommodate 4K or even 8K UD resolutions more efficiently into their cards.
- Also featured on the Australian gaming show Good Game.
- They bugger off the radar again for another year. If you're thinking this is starting to look like a trend, you're not wrong.
2013:
- Not long into the new year, they actually open their new, professional-looking site. Also, their Facebook page, where there's quite a few Geoverse screenshots to mull over.
- Most of the stuff that occurs publicly concerns attending conventions or posting images to Facebook.
- Euclideon puts out a demonstration video for Geoverse, with a small paragraph in the description assuring gamers that they haven't stopped working on gaming stuff.
- They actually release Geoverse sometime around July or earlier. They also put out an interview video with various clients, who basically have nothing but praise for it.
- Then they drop off the radar yet again for yet another year.
- Euclideon spontaneously pops up again in September, this time packaging old Geoverse content as a 'look at what we can do' thing for gamers who haven't been paying attention to Geoverse. In the same video, they also announced that they are opening up a games studio, and someone in the comments with ties to the company claims that there are already two projects in the works that will use the tech.
- They also release another preview vid of similar content with no marketing spiel this time (thank God) a few days later.
- LATEST: TechReport actually manages to get a hold of Bruce Dell, who spills some beans. Yes, there are two games projects, they're aiming for May 2015, they will run entirely on CPU power, there will be animated models (apparently animation is '80% done', and they want the animation to look '100%' first before showing it), they already have implemented dynamic lighting 'better than current engines' but they prefer using pre-baked lighting or real-life lighting due to quality, and they're also working on special anti-aliasing methods. Dell also noted that the platforms being used will be a 'surprise'.
Comment on the latest news: I have no plans on making any predictions on when and what these guys are going to pull something out next, I've tried and failed before. But I have to wonder what kind of game projects they're working on. I know Dell has previously stated that they were working on another playable demonstration with the help of ex-THQ artists, perhaps one of those projects is what they were working on. Who knows. But apparently they're fully focused on gaming content, now that they've finished Geoverse.
I'm personally cautiously optimistic, hoping for the best, bracing for the worst. I really want to see it happen and working, but I'm also worried about being disappointed. Oh, well, that's what happens when you're trying not to get too excited about a seemingly far-fetched but amazing-looking project.