So, what happened to the "Unlimited Detail" graphics engine?

FunnyB 2018-01-02 21:30:16 UTC #1

I was reminded of this today while reading the thread about Star Citizen.

Anyone heard anything about this lately?
I remember my initial thought was that this will probably never work and that it sounded quite fishy.

LexW 2018-01-03 01:44:23 UTC #2

I’m assuming they’re these people:

Given they’re still using that “unlimited detail” tagline and name.

It seems like they’re using the same technology, but have just moved away from game engine stuff. They seem to have real products which people can and have actually bought, but if that’s correct they’re the kind of highly commercial deal that’s only sold via contracts and so on.

Baines 2018-01-03 01:47:15 UTC #3

The idea itself was at least real, of finding efficient ways to render high-detail images using point cloud data. GigaVoxels, for example, dates back to around the same period. (GigaVoxels was supposedly capable of forming realtime images from a dataset consisting of billions of voxels.) As for the “infinite” part, people have been working on various methods for “infinite” detail for decades.

Euclideon has tried to keep its distance from the word “voxels” though, instead seemingly to prefer to describe their system as using 3D “atoms”.

From what I read, while Euclideon hyped Unlimited Detail as a game engine, it worked best with static objects and had issues with stuff like physics and objects that changed shapes. Basically, it turned out to not be that practical for games. (Well, it might have worked really well in walking simulators…)

As for the current state of Euclideon, it still exists. While the Unlimited Detail engine wasn’t suitable for games, Euclideon switched to pushing it for other industries. Euclideon itself has been pushing hologram technology (which uses their Unlimited Detail engine), and last year displayed their prototype for a holographic table that supports up to four users. That table is supposed to go on sale sometime this year, carrying a $47,000 price tag. They are also pushing “Holoverse”, hologram rooms which they call “the next generation of VR”. (They are also still pushing Unlimited Detail engine as an arcade game option.)

FunnyB 2018-01-03 05:44:03 UTC #4

Yeah, that’s pretty much what I gathered as well back in the day. Still, pretty interesting that these guys were reported on everywhere back around when RPS wrote their article, and then pretty much fell off the map. Or at least, no one reported anything else about them.

Baines 2018-01-03 12:23:17 UTC #5

Yeah, even when they first announced Unlimited Detail, I figured it would have major troubles with anything that changed shape.

Hah, I just did a quick web search and found these almost immediately…

http://www.codersnotes.com/notes/euclideon-explained/
Debunking Euclideon’s Unlimited Detail Tech (2016) - A software engineer looked through Euclideon’s patent, and found what he assumes was the backbone of their work. It was a patent based around using an optimized octree, and the guy points out it is basically the same as a 1985 patent for displaying voxels, just with the painting order reversed. He also pointed out that even Euclideon’s demo appeared to be stuck with basic pre-baked lighting for its terrain, and that pre-baked lighting itself tended to be from directly above for instancing reasons.

Euclideon Unlimited Detail - A Fair Comparison (2017) - Euclideon’s own announcement trailer was obviously pretty biased, as polygon graphics at the time already looked a bit better than their polygon “comparison”. This YouTube video is biased in the other direction, but still stands out. It runs some parts of the initial Euclideon trailer, showcasing both their version of polygon graphics as well as their Unlimited Detail tree scene, followed by the Cryengine/Crysis 3 trailer that Crytek revealed only a year later, and then a trailer for the current state of UE4. Next is footage of Euclideon doing their own side-by-side comparison between their Unlimited Detail and Crysis 2, followed by the video maker doing a side-by-side comparison of Unlimited Detail versus Unreal Engine 4.

The Wikipedia page for Euclideon has these details:
Euclideon did sign some big money deals on their Geoverse tech, while they got some government funding for their Holoverse “holographic entertainment center”. Back in 2013, an employ said that they let a bunch of employees go. In 2014, Euclideon said it was working on two games and planned to open a gaming division in the next year.

Shanton 2018-01-09 05:33:43 UTC #6

What ever they are working on , I can’t wait for what the future has! If its legit it sounds like a really cool new innovation

Zephro 2018-01-10 11:54:26 UTC #7

Same as what people were saying at the time? It’s voxel rendering tech, with all its limitations, but with absolute bullshit PR spiel. What they’re actually selling as products are mostly voxel things or behind the scenes product management tools.

nBow 2020-04-03 02:25:41 UTC #8

The idea of smarter software as opposed to stronger hardware is not a new idea. They do base their technology on octree, and I think this is the reason we’ve only seen baked lighting and so far no real physics come out of it. The idea though, of unlimited detail using smarter algorithms as opposed to more processing is not impossible. In fact we’ve already seen working technology like this integrated into common physics and game engines using AI. We have cloth dynamics and particle systems that have proven to be far more efficient simply because they are controlled by an deep learning AI. Big companies do this with their hardware. New Nvidia cards have an AI built in to help with processing. So this concept is not outlandish at all. My personal opinion on this is that even if Euclideon doesn’t solve the problem they have brought light to what we as developers should aim for. We’re at a point now in hardware where we’re almost reaching realistic physics/graphics with standard rendering methods, so taking a logical step and making our software smarter would be a good idea. These types of optimizations to how we render will free up our processing resources for other functionality, and I believe saying this is snake oil or throwing it out all together is not the most intelligent thing you can do. Atom rendering should be where we aim for as engineers and developers, because it unlocks so much possibility for physics in games that we haven’t been able to do before. While I agree that Euclideon is mostly talk and hasn’t had much to show for their claims, we still should try to move in this direction of development. We have search engines in existence that don’t index and can find something from billions of entries, there’s nothing that says building this into searching for pixel information is impossible. Or that optimizing this process is impossible.

LexW 2020-04-03 17:28:59 UTC #9

That implies developers didn’t already know that, which is definitely not true. They didn’t bring light to anything. This sort of thing has been the holy grail for a couple of decades, maybe longer. It’s also weird that you say “we as developers”, because if you’re developer, and you’re in contact with other developers, surely you’re aware that this isn’t a new concept, and has pursued by various people over the years? Unless I guess you’re a solo indie dev or something, in which case fair enough.

Nobody is saying the fundamental concept is outlandish, anyway. They’re saying Euclideon are full of it in claiming they’ve achieved this. If someone did achieve this, Sony, Microsoft, Epic, EA, or Activision-Blizzard or the like (or maybe even Valve, though they might be too proud) would have acquired them in a heartbeat.

I’m rather interested by one of your claims, it’s very specific:

Really? Can you give specific examples of cloth dynamics and particle systems that are controlled by a “deep learning AI”. Industry articles or dev posts would be appreciated. All the cloth simulations and particle systems I can think of in games right now are quite processor/GPU-intensive and not terribly efficient, so that seems like if they were more efficient, that’d be a big deal.

Looking at wikipedia and news articles, it appears Euclideon have essentially abandoned the whole “unlimited detail” thing, despite saying it “wasn’t dead” a couple of years ago, as they’re focused largely on designing and building holographic amusement-park/arcade-type stuff. It is interesting how far holograms have come lately - I saw one in a random bar recently, which was projected on thin air, and which was just blowing my mind (despite being very crummy-looking, just that there was an animated apparently-floating hologram in a random bar - 2020 man…).

Zephro 2020-04-03 18:13:59 UTC #10

I’m not 100% up to date, but definitely not “deep learning” or controlled by an AI. AFAIK from the conferences I was at before the whole Coronavirus thing shut up shop stuff like this: https://www.turing.ac.uk/research/impact-stories/making-simulations-simpler is kinda state of the art. Essentially you’re trying to do a load of finite element analysis (which is fucking hard) quickly, and infamously hard to make parallel effectively. Ah Fortran lessons… So you could use some machine learning to build up a quicker dirtier approximation, and tailor that to say GPU hardware. But once it was baked it’d just be a cloth simulation algorithm so there’s no AI controlling it and it’s certainly not Deep Learning. Though the deep bit is often an overused buzzword as much as it’s anything (more often than not it’s just chucking modern levels of metal at the back propagation algorithm for neutral networks from the 80s ¬_¬ or just trying to sucker investors).

Not that any of this has anything to do with voxels. Which is what the whole thing basically was and no amount of AI is going to deal with the memory bandwidth issues involved. Oh hai there random tree walking on the heap, no my branch prediction cannot help you. You’d basically end up with something like Megatextures in idTech , but yknow without John Carmack doing it.