Judith on 15/5/2020 at 05:21
Quote Posted by Renzatic
edit: A 3.14 million tri sphere with a smiley face sculpted on it gave me an 84 meg .fbx file. I was also surprised at how well Blender handled a polycount that high.
A 4k normalmap compressed with DXT5nm is slightly below 22 megs. 1M poly .fbx mesh is 30 mb. 4M poly mesh is 115 mb.
Sulphur on 15/5/2020 at 05:52
Well, I think we can foresee those 1 TB SSD sizes getting maxed out by around 4 titles, give or take, if we're talking source models and 4/8K textures and lightly compressed audio going directly into a game.
demagogue on 15/5/2020 at 06:45
Next in line is 4+TB SSD becoming standard then. (I want to imagine. I don't really know the technicalities.) But anyway HD size is going to be the next obvious bottleneck if every other game is suddenly shooting for photo realistic fidelity. Maybe cloud gaming could take off to address it. Anyway, somebody will have to do something about it.
Renzatic on 15/5/2020 at 07:06
There are so many variables to account for that it's hard to peg down a guesstimation, but I'd say that your average asset doesn't need more than 1-2 million tris, which will be 30-45 meg, and will only need two large material textures, the diffuse and roughness (and a metallic map, which doesn't have to be 4k, and only needs to be a 1bpp image if used properly, so they can be tiny).
It's hard to say how large the materials will be, since that can vary wildly depending upon the image format. Assuming the best of the best here, we'll say they're using .exr files for maximum quality. I downloaded a megascan to have something solid to work with, and a stack of 4k .exr textures (diffuse, roughness, displacement, and normal) can land you in the ballpark of around 200 meg. Most environment assets are, maybe, around 1500-3000 tris, and for an .fbx file, that's less than a meg. We'll roll with 200 meg for a single contemporary asset for the sake of conversation.
Now a new UE5 asset won't need a displacement or normal map, so we'll cut off half of its texture budget. 100 meg for the diffuse and specular. So you're looking at, all together, maybe 145 meg or so for a single asset?
Keep in mind that I'm only talking about static environment meshes here. Anything that's animated will be, well, probably exactly the same as what we're using now, since even the most miraculously well optimized engine in the world can't make shifting around massive amounts of vertices any less CPU intensive. Character models and whatnot will still be baked.
It's hard to say anything for a fact, but from spitballing, it doesn't seem to make a vast amount of difference, it may even be slightly better.
Starker on 15/5/2020 at 07:15
How many AAA games will have the quality of this tech demo (which was probably polished and optimised up the wazoo) as a baseline, though?
Renzatic on 15/5/2020 at 07:29
To be honest? Probably not too many.
Looking at the export settings in Quixel Bridge, it seems my only other option are uncompressed .jpg files. Downloading the same material in that format gives me a 25 meg stack. Yeah, that's a quite a bit of difference.
Sulphur on 15/5/2020 at 07:53
Epic mentioned that they didn't really optimise it hard or target a high resolution, so that demo's at 30 FPS and (a possibly dynamic resolution) 1440p. And sure, not many AAA games are going to shove in that kind of asset detail in there, but if there's one thing we do know, it's that they'll find a way to use available space anyway.
Renz: think you worked a little too hard on that :D. We're all spitballing here and it could land on either side of anything since we don't have all the variables, but the real constraints are going to be if - as all signs are pointing to - retail is still the marketing priority. Then it comes down to Blu Ray disc capacity and how many they can stuff into those jewel cases. Triple layer BD seems to come out at 100 GB and quad at 128 GB, but I don't think I've seen one of those yet.
Gryzemuis on 15/5/2020 at 10:45
Quote Posted by Starker
How many AAA games will have the quality of this tech demo (which was probably polished and optimised up the wazoo) as a baseline, though?
You ask how many.
Another question you should also ask in this context: when ?
My guess is that all AAA games will have the quality of this tech demo. And cheaper games too. But it might take 10 years. Or more. But imho there is no question that this will be the standard for games at some point in time. And we'll go beyond that. And beyond that. Until we can't make transistors smaller anymore.
When I was a little kid, pong was released. Seeing that was almost magic. How was that possible ?
In my first job (1984), I worked with a computer that did the administration for 3 companies, one pretty large one (an employment agency). 12 People using the system constantly. It was a Data General mini-computer (the size of 3 fridges). It had 512KB ram. It had 2 HDDs, both the size of a fridge too. One was 200MB and the other was 100MB. Having a 1TB SSD in my system doesn't surprise my anymore (I had it for 5 years now). So having a 10 TB or 100TB SSD doesn't seem impossible to me at all. We only have to wait.
Megabyte, Gigabyte, Terrabyte, Petabyte, Exabyte. Who cares ? We'll get there one day.
I'm looking forward to being 80 years old, and playing games with unbelievable eyecandy. :)
Seeing stuff like this UE5 demo makes me happy.
Judith on 15/5/2020 at 11:15
Quote:
It's hard to say how large the materials will be, since that can vary wildly depending upon the image format.
The engine compresses the lossless input you provide on import, and that's usually one of DXT / BC modes. Roughness, metallic, and AO maps are all grayscale; they don't need to be separate textures, so typically they're packed in one image.
All in all, you can actually simplify that to one question really: will the high poly model offset the disk / memory cost of a high-res RGB normalmap?
And the answer is: yes, provided the model is around several hundred K polys, and not around 1M or above.
Starker on 15/5/2020 at 16:48
Quote Posted by Gryzemuis
You ask how many.
Another question you should also ask in this context: when ?
My guess is that all AAA games will have the quality of this tech demo. And cheaper games too. But it might take 10 years. Or more. But imho there is no question that this will be the standard for games at some point in time. And we'll go beyond that. And beyond that. Until we can't make transistors smaller anymore.
Of course. I was not doubting that games will ever get to this point, just the feasibility of having tech of this level in the next generation of AAA games. And sure, I'm as big of a fan of technological progress as anyone. Isn't it absolutely amazing that in just a few decades you go from a 5 MB hard drive that's bigger than an industrial freezer and weighs over a ton to a several TB hard drive that fits inside your phone?
<img src="https://static.techspot.com/images2/trivia/bigimage/2017/2017-03-29-image-5.jpg" width="600"></img>
That said, though, I find that quite a lot of people spend so much time obsessing over whether they could that they never stop to ask whether they should. LA Noire came out nearly a decade ago. How many games are there that use their face tech? Exactly one -- LA Noire. Turns out that not many studios can afford to do what Team Bondi did. Hell, (
https://www.forbes.com/sites/insertcoin/2011/10/05/la-noires-team-bondi-shutting-its-doors/#169ab0181e65) Team Bondi couldn't afford to do what Team Bondi did. Even with one of the richest game companies in the world backing them. And Team Bondi's motion tech might have been an outlier, but it seems that generally the cost of asset creation has absolutely exploded with the advance of more realistic graphics. Top line games now take nearly a decade (or more *cough*Star Citizen*cough*) and hundreds of people to make. Turns out that it doesn't really scale all that well with technological advancements.
And what is it all for? More linear scripted cinematic experiences? More pressure to play it as safe as possible? Huge teams that are more and more removed from the art they create? More perma-crunch and abusive working conditions? Games that have to sell millions upon millions of copies to justify a sequel? Entire genres becoming unprofitable and forced out of the AAA space as expectations and costs go up?
The way I see it, at the end of the day it's a very simple equation -- the more time and resources you need to make a game look good, the less time and resources you'll have to make the rest of it good. I don't think it's a coincidence that two of my all time favourite games (Thief and Dark Souls) already looked incredibly dated when they came out. Personally, though, I'll take that trade-off any day.