The fun part is that HardOCP's staff is so ought of their depths here that they accept Dell's response as a full explanation even though there's basically no new information there. If they had just a little more knowledge they wouldn't have released it at all, realizing what a farce it is, but they're too stupid to see and too arrogant to ask someone more knowledgeable's oninion.
Unlimited Detail, octree traversals
- Thread starter Alucardx23
- Start date
Yes, there's a 3 year old thread that's also covering this where he posted. He hasn't addressed any of the complaints though. He's engaged in no technical discoussion whatsoever. He amkes some wild claims that fairly get questioned but just fobs off the questions with broad "we're doing something new" statements, while simultaneously admitting he can't know if it's new or not because he works in isolation and hasn't seen what's been tried before! Being a 'nice guy' isn't enough to deserve people's trust in a technical matter, especially where there's investment involved. And it's to be expected that when he ignores legitimate queries, he's just going to generate mistrust and 'snake-oil' type comments. He has brought this ruckus on himself by not being open. He doesn't ahve to give away trade secrets, but he should at least explain what issues he's solving and give some evidence that they those problems he hasn't demo'd soltuions to can be resolved.
I don't mean to defend his (lack of) answers to legitimate questions, but I think he means well and is truly oblivious to advancements made by others in this field. And I think he may not fully grasp the shortcomings of this technology when it is implemented in a real game.
Right. I concur. Problem is he's not willing to listen. Many more learned people than Dell have identified issues and he's just ignoring them, instead of listening and proving he has a workaround (necessary to have a legitimiate claim on people's investment) or realising it isn't the solution he imagines it to be. Up until now I'd be saying, "good for you," but now I've got to the point where I'm thinking he needs to change to be more professional about this and what he's doing.
The investors are an important point.. if I were them, I'd be somewhat concerned by all of the online backlash, it would make me think that any money I put into this project could very well be wasted in the end. After all, the investment will only pay off if people actually license the engine and make games with it.
And from the Australian government's point of view they're probably not concerned about 2 mill, he's hiring 8 people in the 'tech industry'. It probably doesn't matter too much to them if the product is practical or not as long as there are people actively researching things.
Even if the project results in untextured, badly lit environments with everything facing in the same direction and consuming lots of memory (assuming you want to do anything more than move the camera around a static scene) - he's at least skilling up some of the workforce. It's hard to imagine it's gone on this long without him doing a literature search but you kind of get the impression he doesn't know about any of the prior art. If you believed you were the first person to discover sparse voxel octtrees you'd be pretty chuffed wouldn't you? Especially when you keep getting funding.
Tbh, I'm starting to wonder how long it would take to replicate one of his demos and open source it without all the PR spin. Just a list of what it does, what it might be useful for, and what its limitations are.
Even if the project results in untextured, badly lit environments with everything facing in the same direction and consuming lots of memory (assuming you want to do anything more than move the camera around a static scene) - he's at least skilling up some of the workforce. It's hard to imagine it's gone on this long without him doing a literature search but you kind of get the impression he doesn't know about any of the prior art. If you believed you were the first person to discover sparse voxel octtrees you'd be pretty chuffed wouldn't you? Especially when you keep getting funding.
Tbh, I'm starting to wonder how long it would take to replicate one of his demos and open source it without all the PR spin. Just a list of what it does, what it might be useful for, and what its limitations are.
I was thinknig exactly the same thing! If I was an undergrad, I'd think about riding this popularity. The only thing about the demo that's impressing people is the size of the dataset, and the detail he's captured. Replicate that, stick up a movie, and list the technical limitations. It's worth noting other demos of voxel tech have included lighting/shadowing, so Unlimited Detail can even be trumped.
digitalwanderer
Legend
Yeah - it seems like it's a perfect project for something like a masters degree final project. Most of what's been shown by unlimited detail should be pretty straight forward to code up.
The old demos with the same objects repeated many many times in regular pyramid patterns can be done very simply (even if Bruce Dell's claim that it isn't just a big octtree is true) the end result would be identical. It's not that big a jump to move away from creating an object at every node and end up with a 'game world' as shown in his latest videos. Especially as there's no requirement to texture objects differently or change their alignment.
Gut feeling says that Euclidean has some kind of special sauce to allow them to address the data quickly from the camera position that isn't going to be trivial to do. It's likely just trigonometry or a method from the ray tracing literature but it's likely inventive enough to justify a masters degree. It would be an interesting project to code up... Plus it leads into ray casting, ray tracing, lighting and texturing which are all the things we as an industry care about. Would be a very employable person when they came out the other end of it.
To play devil's advocate here, assuming it's an SVO, can't they do spatial colour compression here? I mean, they only really need a full colour value once in a while as many of those [strike]nodes[/strike] ATOMS next to each other will likely share the same colour.
It'd make the [strike]traverse algorithm[/strike] FETCHING ATOMS more complex as you'd need to keep going down level until you completed the 24 bits of colour information but you could have considerable savings in the end. Or you could just interpolate the colours from peer nodes if you didn't care for the accurate colour or really wanted a clean depth-limited traversing.
From users, yes, YouTube comments and the like.. but the reaction from their potential customers, the game developers, has been pretty iffy. It doesn't matter if everybody else likes it if the people who will actually use it don't.
I disagree somewhat. People are already demanding that Epic should immediately buy this tech and include it in Unreal Engine 4. All in all it can generate bad publicity, as it's hopeless to try to explain the drawbacks and problems to the general audience, and they're already emotionally commited to the support of Euclideon.
Of course an eventual backlash seems almost inevitable, too, but that'll take years, and Dell would have to release another incomplete demo with the same magical claims. Who knows, if he's really just this ignorant and not trying to scam people, it may eventually happen...
Of course an eventual backlash seems almost inevitable, too, but that'll take years, and Dell would have to release another incomplete demo with the same magical claims. Who knows, if he's really just this ignorant and not trying to scam people, it may eventually happen...
Ok, everyone just take 3 steps back and breathe deeply.
Then remember what the internet thought of OnLive, and how people who are the absolute experts of their field (DigitalFoundry, etc) swore up and down that this tech was *impossible*. Not just hard to pull off, but IMPOSSIBLE. And yet it came to be, and it works quite well.
Now we have this Bruce Dell who has very few preconceptions of what is supposed to be possible or impossible regarding voxels. Obviously he is not stupid, because he is capable of programming his own 3D engine, which puts him above 99,99% of the commentators here.
So just consider the possibility that a keen mind not bound by conventions has found a novel way to surpass the limitations HE DID NOT KNOW EXISTED.
And stop with the detracting already, based on "he's not telling us anything specific enough about the technology". Having found a (probably quite elegant and simple) way to solve a problem previously considered unsolvable, and standing to make a lot of money off it if it came to fruition before copycats ran rampant, who would?
Also, the hardware that runs the preliminary unoptimized demo is NOT what you would call high-end - it's a *laptop* processor (i7 2630QM 4C/8T @ 2GHz). If you wanted high-end, you would see it running on 6-core 12-thread i7 990X @ 3,46GHz.
My take is that the system was held back by something else (probably unoptimized software or by amount of RAM in the very extreme case), and that the CPU was not the bottleneck in this scenario.
Then remember what the internet thought of OnLive, and how people who are the absolute experts of their field (DigitalFoundry, etc) swore up and down that this tech was *impossible*. Not just hard to pull off, but IMPOSSIBLE. And yet it came to be, and it works quite well.
Now we have this Bruce Dell who has very few preconceptions of what is supposed to be possible or impossible regarding voxels. Obviously he is not stupid, because he is capable of programming his own 3D engine, which puts him above 99,99% of the commentators here.
So just consider the possibility that a keen mind not bound by conventions has found a novel way to surpass the limitations HE DID NOT KNOW EXISTED.
And stop with the detracting already, based on "he's not telling us anything specific enough about the technology". Having found a (probably quite elegant and simple) way to solve a problem previously considered unsolvable, and standing to make a lot of money off it if it came to fruition before copycats ran rampant, who would?
Also, the hardware that runs the preliminary unoptimized demo is NOT what you would call high-end - it's a *laptop* processor (i7 2630QM 4C/8T @ 2GHz). If you wanted high-end, you would see it running on 6-core 12-thread i7 990X @ 3,46GHz.
My take is that the system was held back by something else (probably unoptimized software or by amount of RAM in the very extreme case), and that the CPU was not the bottleneck in this scenario.
I'm a big advocate of new ways of thinking, and I frequently mention that clever people find clever solutions and I expect that to continue. For a while I was willing to afford this guy a little optimism, and if he does pull it off, good for him. Refusing to discuss the potential issues he faces and potential workarounds is contrary to the while idea of mature discussion though. Okay, perhaps some reactions are pretty emotional ("he's a con artist") but at the same time there are con-artists who bamboozle people (Phantom) and trying to protect people from getting bamboozled is a Good Thing.
So IMO examining what he's doing, identifying the problems, and then questioning what's really possible and what he can pull off, is one of the most useful and practical things a 3D discussion board can do! Unlike OnLive who have taken experts in their fields and talked through what they are doing and the tech they created and how it's addressed a limited set of problems (compression time and efficiency, and latency, which are still issues that precent OnLive replacing the console experience as some detractors were saying), Dell has just said he can do everything (lighting, shadowing, animation) without demoing it. And without using raytracing either, which is perhaps the ideal solution for integrating voxels with lighting and shadowing. The positions are not at all equivalent.
Did people here really believe that OnLive was impossible? Client server models are pretty well known... I recall people saying that the latency would be unacceptable (which for many cases it still is) and that it's not a very efficient way of sending data around the internet and that's why most games (e.g. MMO's) retain local rendering and send small data packets that synchronize events. Isn't this still the case?
I think you dramatically underestimate the skill set of your audience.
I don't think people are that opposed to the idea of new inventions - in fact as an industry we constantly require those ideas to be generated and adopt them incredibly quickly. The issue comes when 8 years pass with no further information and no progress in overcoming or even acknowledging the issues that are well known. It's not like he's pioneering some unknown and unexplored part of graphics that hasn't occurred to anyone else. He's clearly using sparse voxel oct trees and generating hundreds of the same object. Sparse voxel oct trees and instancing IS NOT NEW. I don't know how mnay more people can explain this.
To add to what Shifty said, it's also a matter of claims versus proof. OnLive actually does not work according to their original claims, that latency wasn't a problem (it's noticeable - some games more than others) and that they'd be streaming 720p (and above!) in real-time: there are noticeable video compression artefacts. It's impossible doing what they claimed it would do.
Same here with Euclidean. The claims are pretty outrageous. The unlimited part is demonstrably false because there isn't any computer with unlimited HDD to store such an unlimited detail. The only way to do unlimited detail without requiring unlimited storage is using fractals and similar mathmatical procedural generation which very few games have used, despite this being the de facto strategy in 4K demos. Just because you can make a tech-demo does not mean you can make a game engine.
I also agree with Shifty that people have gone too far. Notch in particular should be careful, his bank account is big enough to atract a libel suit. Calling someone a scammer on their personal blog is subject to this kind of thing. However your argument that 99.99% people here shouldn't comment because never wrote a 3D engine is severely narrowminded. I've written three 3D engines from scratch and I'm not even a graphics programmer; nowadays it's not a big deal. There's a lot of people here writing their masters or doctoral thesis and know a thing or two about how the scientific method works.
If you make a claim, especially an outrageous claim, it's expected that you provide proof. If you don't provide proof you're going to be either ignored or exposed. This recent interview is the first indepentent proof we've had in 6 years: seeing something run real-time on known hardware. Btw, if you don't think that gaming laptop (Bruce's words) isn't high-end you're mistaking high-end with highest-end. That laptop has the top mobile cpu and 8Gb of ram. What's not high-end about it? Yes, it's not a desktop machine and it doesn't have a $999 CPU. It's just a high-end gaming laptop.
Also, again with the claims, this technology is supposed to work on current gen consoles. Nevermind the fact that both will probably be replaced before this technology is released but when that laptop runs at 15-25fps at 1024 x 768, with huge amount of duplicate objects it doesn't inspire that much confidence.
As for the "he shouldn't provide details otherwise people will copy it", actually publishing something will establish prior art and allow him to apply for a software patent (ugg - I can't believe I'm writing this). Keeping it a secret is actually more dangerous because patents don't have to be performant: someone could use a clumsy unoptimised version of Bruce's algorithm, quickly patent it and then Bruce would have to pay license fees.
Same here with Euclidean. The claims are pretty outrageous. The unlimited part is demonstrably false because there isn't any computer with unlimited HDD to store such an unlimited detail. The only way to do unlimited detail without requiring unlimited storage is using fractals and similar mathmatical procedural generation which very few games have used, despite this being the de facto strategy in 4K demos. Just because you can make a tech-demo does not mean you can make a game engine.
I also agree with Shifty that people have gone too far. Notch in particular should be careful, his bank account is big enough to atract a libel suit. Calling someone a scammer on their personal blog is subject to this kind of thing. However your argument that 99.99% people here shouldn't comment because never wrote a 3D engine is severely narrowminded. I've written three 3D engines from scratch and I'm not even a graphics programmer; nowadays it's not a big deal. There's a lot of people here writing their masters or doctoral thesis and know a thing or two about how the scientific method works.
If you make a claim, especially an outrageous claim, it's expected that you provide proof. If you don't provide proof you're going to be either ignored or exposed. This recent interview is the first indepentent proof we've had in 6 years: seeing something run real-time on known hardware. Btw, if you don't think that gaming laptop (Bruce's words) isn't high-end you're mistaking high-end with highest-end. That laptop has the top mobile cpu and 8Gb of ram. What's not high-end about it? Yes, it's not a desktop machine and it doesn't have a $999 CPU. It's just a high-end gaming laptop.
Also, again with the claims, this technology is supposed to work on current gen consoles. Nevermind the fact that both will probably be replaced before this technology is released but when that laptop runs at 15-25fps at 1024 x 768, with huge amount of duplicate objects it doesn't inspire that much confidence.
As for the "he shouldn't provide details otherwise people will copy it", actually publishing something will establish prior art and allow him to apply for a software patent (ugg - I can't believe I'm writing this). Keeping it a secret is actually more dangerous because patents don't have to be performant: someone could use a clumsy unoptimised version of Bruce's algorithm, quickly patent it and then Bruce would have to pay license fees.
We can't be sure he's using SVO exactly even if the results are remarkably similar. He may have his own data organisation system. It's the fundamental principles of volume-point based data that are raising concerns. If every point can be moved and changed (animateable) than you can't rely on obvious datastructures to represent that data, so compression goes out the window.
Take Richard's post above (prior to his reply while I was typing this!). If every colour sample is a point, then spacial compression as he suggests is an option, but then raises the question of how you can include vairable textures - think blast damage decals or user-created decals on a car or something. The moment you have arbritrary access to any colour sample, you limit the available compression options. Textures get around this nicely, hence they have been used in all 3D industries for a long time, and no-one like Pixar is planning on going with offline renders created with gazillions of individually coloured points. the datasets are unmanageable. Megatexture and megamesh are having to address these issues of huge datasets in simpler contexts to the incredible data density of voxels.
If Dell has found a way to solve this issues that noone else has thought of (which does happen), then he hasn't shown anything to demonstrate that. His current and all previous examples are just lots of the same tiles/objects. He could easily use a fractal program to generate that much unique data to explore, to show streaming variety isn't an issue, but he hasn't done so.
Take Richard's post above (prior to his reply while I was typing this!). If every colour sample is a point, then spacial compression as he suggests is an option, but then raises the question of how you can include vairable textures - think blast damage decals or user-created decals on a car or something. The moment you have arbritrary access to any colour sample, you limit the available compression options. Textures get around this nicely, hence they have been used in all 3D industries for a long time, and no-one like Pixar is planning on going with offline renders created with gazillions of individually coloured points. the datasets are unmanageable. Megatexture and megamesh are having to address these issues of huge datasets in simpler contexts to the incredible data density of voxels.
If Dell has found a way to solve this issues that noone else has thought of (which does happen), then he hasn't shown anything to demonstrate that. His current and all previous examples are just lots of the same tiles/objects. He could easily use a fractal program to generate that much unique data to explore, to show streaming variety isn't an issue, but he hasn't done so.
Similar threads
- Locked
- Deleted member 13524
- PC Hardware, Software and Displays
- Deleted member 2197
- General Discussion