Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

A first look at Unreal Engine 5

unrealengine.com

Post reply on HN
Print view View original

351–360 of 695 posts

Re: A first look at Unreal Engine 5

#351

Post by 6gvONxR4sf7o »

xxs wrote: post #347
6gvONxR4sf7o wrote: post #209

Earlier quoted context omitted.

Don't you still need to do some of that just for file size and data transfer, if nothing else? In the demo they said it had hundreds of billions of triangles. Google tells me you can get down to 18 bytes or so per triangle, but even if we assume a lower bound of just three bytes per triangle (amortized storage of 1 vertex per triangle of an x,y,z point), we're talking about on the order of terabytes to just store tha…

>even if we assume just three bytes per triangle, that give's you up to 16.7m (2^24) possible triangles. Your napkin math is way off,

I edited it to clarify I meant that as a lower bound. Even if it were somehow achievable, 3 bytes per triangle times hundreds (plural, so at least 200) of billions of triangles is at least 600 billion bytes: 600 gigabytes.

Re: A first look at Unreal Engine 5

#352

Post by moultano »

m12k wrote: post #292
gautamcgoel wrote: post #244

What a magical demo! That made my day. I had largely written off console gaming as a never-ending sequence of brain-dead first person shooters. I'm glad to see real innovation in the graphics space - coupled with advances in AI and NLP, I think we could see a crop of incredibly realistic and meaningful games in the next few years. IMHO, a key innovation that still needs to occur is a shift away from flunky and unintu…

To play devil's advocate - games with this kind of production value require a huge investment in asset production, which means they tend to be very conservative when it comes to gameplay, because they want to be damn sure it doesn't flop. So until the production of something like this comes way down in price, the games that look like this are likely to be clones of existing successful games, e.g. Uncharted, brain-dea…

Maybe, but photogrammetry is a waaaay cheaper way of producing assets, and being able to directly ingest high res models from a scan could actually drop the art costs of games a great deal.

Re: A first look at Unreal Engine 5

#353

Post by 6gvONxR4sf7o »

omikun wrote: post #276
6gvONxR4sf7o wrote: post #209

Earlier quoted context omitted.

Don't you still need to do some of that just for file size and data transfer, if nothing else? In the demo they said it had hundreds of billions of triangles. Google tells me you can get down to 18 bytes or so per triangle, but even if we assume a lower bound of just three bytes per triangle (amortized storage of 1 vertex per triangle of an x,y,z point), we're talking about on the order of terabytes to just store tha…

That's assuming 6 bytes per vertex, 3 vertices per triangle? Most assets use triangle strips, which averages closer to 1 vertex per triangle. They also said the souce mesh is in billions of triangles, but their Nanite pairs that down to 10's of millions. At ~1 vert/tri * 6bytes * 100mil tri, that's 600MB of meshes. Keep in mind this is just a demo, so they can pack what would normally be an entire level into one room…

I was going off of this:

> Thus, when vertices are shared, the total amortized storage required per triangle will be 12 bytes of memory for the offsets (at 4 bytes for three 32-bit integer offsets) plus half of the storage for one vertex—6 bytes, assuming three 4-byte floats are used to store the vertex position—for a total of 18 bytes per triangle.

From here: http://www.pbr-book.org/3ed-2018/Shapes/Triangle_Meshes.html

And I they said hundreds of billions of triangle, not just billions (timestamp 6:29), but another commenter made a point that's obvious in retrospect that it's a ton of reused assets, not hundreds of billions of totally independent triangles.

Re: A first look at Unreal Engine 5

#354

Post by quotemstr »

muststopmyths wrote: post #240

This looks nice, but I'm not sure what the practical gameplay consequences for the new tech are. I sure as shit am not going to ship a terabyte of original-quality assets in my game. It would help me produce cinematics more easily though, I suppose. Real-time global illumination looks dope. No more overnight lighting builds :)

If I understand the geometry images paper correctly, you don't have to ship the original assets. The technique uses fancy topology to encode a mesh as an image --- x, y, z of each vertex being the r, g, and b channels of the image --- and pages the images into the GPU on demand. You can downsample the images as much as you want

Re: A first look at Unreal Engine 5

#355

Post by zokier »

klmadfejno wrote: post #78

Can someone help me articulate my issue with this? Looking at a lot objects, they seem unnaturally detailed. Like the bugs, or the statue at around 6 minutes. Something about it makes me really uncomfortable feeling physically. Is this like a weird infinite depth of field kind of thing?

There are all sorts of weird (temporal?) artifacts in the demo, I think especially visible in the climbing scene. Look at this cropped capture for example:

https://ibb.co/PrYT6pq

The region below her arm is much sharper/detailed (missing dof/motion blur?) than surrounding area. It seems to be somehow related how stuff is revealed, as her arm was moving up in this bit, so the overly sharp region was covered in previous frames.

Also the shadow (e.g. her fingers) has very sharp pixelated edges which can contribute to the feeling of oversharpness. There are some other lesser artifacts also abound, so yeah, its obvious that the engine cuts corners. But of course that is to be expected, realtime graphics is all about compromises.

Re: A first look at Unreal Engine 5

#356

Post by haywirez »

neotriple wrote: post #98
aeturnum wrote: post #34

Earlier quoted context omitted.

Yah, they said "convolution reverb," so I think they expected audience to know what it was. Often it's not about creating new technology, but making existing technology accessible to new audiences and new creators.

Honestly, convolution reverb is already highly accessible. It’s a feature in $50 (and probably free) reverb plug-ins for anybody that does audio. They’re relatively easy to implement and have probably been around for a decade+. As the OP said, the audio things they mentioned aren’t new or adding any accessibility imo

You have it right now in the browser Web Audio API, for free[1]

[1]: https://developer.mozilla.org/en-US/docs/Web/API/ConvolverNo...

Re: A first look at Unreal Engine 5

#357

Post by yodon »

VikingCoder wrote: post #327
yodon wrote: post #301

Earlier quoted context omitted.

The hard part of the multiverse is the business model. Features like dynamic maps are just that, features. If you have a business model that works you can bring all the tech features to play. If you bring the tech to play without a business model you just crater spectacularly. This is why Fortnite is important. It may not be the feature list you thought the multiverse would launch with but in hindsight it will be the…

For me, if the content is compelling, I'm paying a monthly membership. I expect the multiverse to pay content creators based on how much time I linger in their content. Did HTML / HTTP have a business model? Or were they a platform upon which other things were built?

The business model for HTML/HTTP was and is advertising.

People initially tried subscription models for HTML content but it didn't work so everyone went to advertising based models. Observationally, subscription HTML content models have continued to not work (yes, Netflix, but Netflix is more about video content than HTML content).

Re: A first look at Unreal Engine 5

#359

Post by xxs »

ben7799 wrote: post #299

It mostly looks amazing but also very uncanny valley. The development process improvements sound most promising. I was amazed until the the character started climbing the rocks. Seems like no one at Epic has ever climbed a rock in their life.. it sent the whole demo back to PS1 land. It doesn't even look like a super hero rock climbing, just looks totally wrong. Almost no part of the body ever looks angled correctly…

Similar feeling but starting off with the initial leap. The animation of the interaction/grab with the cliff edge looked very off - climbing without the feet touching the rocks.

Creating a scenery with a lot of details depth is fine but it doesn't even come close to having proper animation/character models and especially realistic physics.

Re: A first look at Unreal Engine 5

#360

Post by vasili111 »

All this for sure looks amazing. But I do not think it will bring anything new to gameplay. There is no big gameplay changes in past 15 years. All shooters, all MMORPG, all strategy games and etc are based on the ideas of 2003-2005 years. Graphics are improving but most games have similar gameplay. I think the biggest next impact to gameplay and game industry will be engine that will be able to have huge zoom ability + ability to handle more than 30 000 players on the same map (of course that also depend on server software and not only game engine). Like traveling from space on ship to planet and after landing and play as a regular 1st person game but all of this with lots of players on the same world (more than 30 000). I think that is next big thing that will change game industry.
Post reply on HN