Unreal Engine 5 = Game Changer

If they truly deliver everything on the demo i might start migrating to Unreal, with all the cool new tech and license changes we may start to see more jobs for Unreal.

 

Q1: How much of the demo is the new Unreal engine and how much PS5 hardware?

For instance the instant GI could just be AMD RayTracing.

And pushing that many triangles surely is only possible with a powerful GPU.

Q2: What about being able to stream that amount of data in real time, is the bandwidth between IO/CPU/GPU so high in PS5 that it can throw around way more complex scenery data?

Q3: Maybe the test would be to see the game running in editor on a desktop PC and knowing the hardware specs of the PC?

 

Same, I'll wait until UE5 is out to make this huge choice.

 

Man that dynamic GI, and on AMD chipset which means no need for those silly NVIDIA tensors.

 

So much power. I wonder if it will create anything I want to play thoug.

 

I fully expect it to be using raytracing hardware to accelerate itself in some way because it just makes sense due to the nature of the current implementation of raytracing hardware. NVIDIA's implementation is currently just hardware that performs optimized raycasts (they check for a bounding volume and then for a triangle).

Tomorrow at 6 AM Pacific is the NVIDIA GTC announcement stream for Ampere GPUs. One of the current rumors floating around is that the performance increase will be sufficient that NVIDIA will have RTX across their entire product stack and that traditional GTX cards will be completely obsoleted. I'm hearing up to quadruple the raytracing performance.

 

Learning new game engines is similar to learning new programming languages. Your first one will always take the longest but future ones will be progressively faster to learn.

 

Apparently the demo is not using ray tracing.

 

Yes Ps5 IO is 5.5GB/s (Raw), Typical 8-9GB/s (Compressed)
So yea 8-9GB/s compressed is very fast

 

Yeah, I know that feeling. Both the large collection and the impulse buying. That said I haven't had too much trouble getting artwork from the Unity store into UE projects, and some tools exist for both engines so you might try contacting the author and seeing if they support it.

 

I cannot even look down on Unity for being so far behind.

This is so disruptive compared to everything we have been doing up until now I believe the entire entertainment industry must be shaken by this, even the film previz one.

I am still sitting down trying to process all this and how to move on with my goals based on this.

This came out of nowhere and changes everything.

To put it simple: Direct Zbrush exports... like... like... oh boy. It means the focus is no longer on retopology but rather an animation tool able to work with Zbrush meshes without chugging.

My jaw is on the floor.

If I can do everything in Zbrush, then even I could make a AAA game alone without even exaggerating and I'd go as far as saying anyone with a decade of experience will also be a able to.
The reason, people, is that 'Zbrush assets' never get outdated.

The reason I am sitting down to recover is because this reveals I have been wasting a great part of my time in the last two years working so much on tools and pipelines to simplify production that suddenly this all turns a great part of my work into pure timewaste.

EDIT: Glad to know I am not the only one.
Every developer I know is F***ing losing their minds in my chat group.

 

Thought #1: Great, now I need to update skeletal mesh import again.
Thought #2: And now I can make my lowpoly game with realtime gi I always wanted.
Thought #3: What's wrong with her nose?
--------

This seems to be geared towards film-quality assets. It also probably takes insane amount of disk space.

As long as those systems work with ultra low poly assets, however, this is good.

 

Well, my 2 cents are all about skepticism:
1)Real time GI looks cool, but probably produces tons of distance-based optimization artifacts that will be very visible on clean white materials and flat geometry (architecture). With rough nature surfaces it's not visible, but on clean flat surfaces i bet there will be a nasty surprise...
2)Same for auto-LODs - with nature assets like rocks those auto-lods may be not noticeable, but tech and architecture models will have a lot visual problems with that I presume.
3)Unity has much better implementation of ray tracing (2-3 times faster than Unreal, drastically better noise filtering, etc) - probably it's possible to create 95% of this Unreal 5 demo in Unity HDRP + DXR and not in 2021+ but right now, using traditional LODs technique.
4)Epic games has a history with showing "super realtime GI" in promo materials for engine and cutting it off in real engine: VXGI didn't arrived) -

5)Those words about "it will be easy to migrate from UE4 project to UE5" are party killers for me (seems that there will be not so much changes, still polished but very old Unreal engine 4 inside).

So, in the end it's some reverberation zones improvements (not even ray traced sound propagation it seems) + useful, but limited to several types of assets new auto-LOD and very good streaming of models/textures/etc - and all of this on top of clumsy old UE4.

Unity's ECS promising Megacity-like projects (200K+ draw calls, also new streaming for assets) - big programming work was done and still going, that's feels like much better and solid foundation for future growth.

 

Does the polycount not just mean the PS5 has a very powerful GPU, what about occlusion culling tech?

 

The unnamed thing here is the new console architectured around SSD, in particular sony virtualizing RAM to SSD, filling 16gb in 2s (150mo per frame at 60) is a real game changer, the xbox is only 2/3 away though in raw bandwidth but doesn't have the 16 distinct data channels. That's why it's a ps5 demo and not a pc demo.

As soon as I have seen unlimited details, I was like "they put the SSD to good use" I fully expected a sort of mesh streaming solution that take level and optimize the polygon soup directly instead of staying per mesh. This is something I have been thinking for while, but I have not the technical chop to pull it, but anyway, that's how some big AAA engine already does it, I know the DOOM eternal does it, and the Rainbow six engine does it too, along with virtualized texture, it was to be expected, on top of meshlet shader, we were due for some big change in that department. Obviously this mean refined occlusion culling.

Being able to stream and swap fast will help make visual you would never see otherwise, by simply brute forcing and hot swapping. Also the mandalorian effects was made with unreal, obviously movie production is shaking right now, most notably small shop will with excitement. With disney being currently in a dire situation (they have raise 25 billion in loan to stay afloats, negating the entirety of 2019, with rumor they were in bad shape despite those earnings as the parks are still their best revenue with 1/3 of all earnings but now close due to covid and still need maintenance) and the crew members scattered with furlough and unemployment, along with the faltering comic industry, we might see some surprise contender jumping on that opportunity to create new model. We have seen the raise of single man animation (I think a warhammer fan film was done by a single guy), I expect this to consolidate with the great confinement (number of devlog recently started on youtube skyrocketed).

 

I'm very skeptical about ECS, to be honest, and promising does not mean it is already there. The whole ECS/DOTS things gives me same vibe as situation when a programmer gets fascinated with a certain algorithm and tries to apply it everywhere.

 

I am a huge Unity fan...but let's talk facts...

I am not going to compare to legacy or URP as that would be insulting...
I will just talk HDRP vs UE5
1) HDRP has no Realtime GI...(deprecating...only available up to 19.3) I don't care at this point when Unity doesn't have the feature at all.
2) auto LODs...are not in Unity and the existing ones are really crappy, breaking with lightmaps, breaking with culling etc.
3) Nope, u got it wrong big time. Here is why. The poly count on UE5 demo is using Hi poly assets from quixel. That's like 500k triangles for a small piece of rock. Filling a whole scene (even if it was some trickery of visualization) full of those assets is not possible on Unity. In fact, UE5 is probably the first engine to do so. Heck, it will be impossible to work with them in the editor to begin with...
4) When it comes to bait and switching through video marketing, I think Unity tops Epic...I agree with you that Epic is probably overselling it, but remember that UE4 already is miles ahead in terms of rendering...I can't imagine UE5 being a regression...just doesn't make any sense.
5) Migration is pretty standard practice nowadays...really not a sign of anything.

So, my gut feeling + rational prediction

UE5 is pretty much free now until 1 mill in revenue.
UE5 has realtime GI
UE5 has auto LOD - that allows us to skip on Normal maps and LOD generation. Also leaves us free from poly count restrictions...
UE5 has a full photogrammetry suite for free (quixel megascans)

So...UE5 is free until 1mil.....
No need to bake GI
No need to bake Normal maps
No need to retopologize
Has a huge library of assets for free...
but no c# support...

damn...

 

Also unity is on a bad endless treadmill of unfinished content, they promised tomorrow features that are deprecated today for a new shiny.
Unreal, here is the fully developed and tested solution, we make games thrust us, it will fit.

 

I don't get it. To achieve such impressive level of graphics you need to spend millions of dollars of hi-end assets, animation, sound and crew. Indie developers can't afford it. Companies that can, like EA or Ubisoft, have their own proprietary engines. I feel that outside of film industry, there are very few developers that can really use UE on full potential.

And what are those AAA UE4 games that everybody keep mentioning? I only know of Gears 4-5 and Days Gone, the rest are just mediocre.

P.S I am almost sure that majority of people that keep saying they are switching to UE are hobbyist that newer released a single game.

 

Small prints: on ps5 and probably new console due to ssd optimization that remove bottleneck.

 

Speaking of which, do you rememebr blacksmith demo? It did look great (aside from that song which didn't really fit), but when it was released it become obvious that there's ridiculous amount of custom scripting and tweaks in there.

The demo, while didn't explicitly say it, implied that "you will be able to do that too, easily". Which wasn't true.

On related note, it would be nice to see how all of the stuff in demo work on whiteroom and on super low poly assets.
This is all very promising, but it is most useful for super high budget titles, and also likely makes installation size balloon. Unreal 4 games already can require 80 gigabytes of disk space. If they store photogrammetry-level data, this requirement could go up to 250 gigabytes or more.

 

Like, seriously.

How are some people dismissing this lol!

Unreal Engine 4 is already the top engine when it comes to graphics quality output and has had that quality since at least 2014 or so.

And people are surprised that Unreal Engine 5, 6 years later, can do this?

They haven't been slacking for all those years.

 

When your package manager has more preview icons than not, and the blog is all AEC oriented with random company acquisitions, you begin to question.

Even so, there's a couple more years before that wonderful tech demo becomes fruit for everyone to play with, and in those 2 years I can't help but wonder if my preview package count will go up or down.

 

Any demos opening up this project in the editor? I am skeptical and interested in seeing how this works outside a "tech demo". Until we see how easy this ZBrush 8K texture project is to throw together I will remain skeptical.

 

I'bet those are not real polygons but something like this
https://www.extremetech.com/gaming/91854-unlimited-detail’s-non-polygon-3d-gets-more-unlimited
https://venturebeat.com/2018/04/05/atomontage-reveals-voxel-3d-simulation-technology-for-games/

still very impressive though, the animation tech too

 

Use your head: that video shows the ceiling.

If you use lower quality Zbrush assets you can get away with much bigger game worlds and you still get free GI.

If you are an Indie then you must think like an Indie dev how those tools greatly benefit your workflow. It isn't suddenly demanding you to make 'high quality assets' but rather 'now there is much less need to optimize when you are going for a scene with way less complexity'.

But you still get free GI.

Free GI.

 

I can certainly vouch for the performance gains across the board, even for smaller projects, but the workflow for actually doing things in ECS is absolutely a mess.

 

They already explained what the technology is all about: realtime lossless reduction of high poly sculpts.

So no. It isn't all fake smokes and mirrors like Euclideon. Epic has a name and a reputation and they aren't going to stain it with false advertisement.

Some of you should try harder. You are coming across as trolls.

EDIT: The statues might be instancing just like Euclideon (and Unity Megacity) but none of the initial high poly geo is.

 

NO

they explicitly talk about triangles, those tech are non triangles.

I wonder if they have improved GPU dealing with small triangles, which was a big bottleneck in teh pst due to the 4x4 pixels compute.

 

Also Valorent is another testament of unreal on low end hardware, let's not forget the other end of the spectrum

 

I also like how they actually put some gameplay in there. Rare for a demo!

 

Dynamic global illumination should work fine on low and medium poly meshes. That's great for game devs. The ZBrush -> GoZ -> Unreal stuff ... we'll see.
Animation topology being my top point for scepticism here - there simply are situations were retopology is not avoidable. For environment art that nearly unlimited polycount is great, especially for targeting the film industry which UE5 is definitely going to do (and has been doing successfully in the past years). Imagine a director telling an artist to change some rocks on the fly while being at the set and not having to wait for decimated and unwrapped meshes. They could use vertex colors at that level of detail.

 

That's the thing.
If Unreal Engine can run that in realtime, now the pressure is on dcc tools providers to come up with a rigging/animation system that works on high poly mesh taking advantage of modern processors and GPUs.

This is the kind of disruptive event that forces the industry to push forward in order to keep up.

There is a reason none of us even considered something like this: because 'current hardware architecture doesn't make it feasible'.
But Epic just did this. The impossible. And now we are all doubting if this isn't just some fake advertising because if it isn't... as you say 'they could use vertex colors at that level of detail'.
A seemingly innocuous feature that suddenly opens up the world to a totally different approach to how we are doing things.

Some of the gamedev contacts in my chat group are saying this all feels like a bad dream lol. It may not be as dramatic as it sounds...

 

Damn. Dat look good.

 

And that's why I'm skeptical. There's a reason why we, for example, aren't writing things in assembly while trying to maximize performance.

 

I'd say 'how it looks' has got nothing on the workflow in terms of being impressive.

 

The chances of me creating art that can utilize any of these features are basically zero. But I have to say, I look at this "Royalty-free until $1 million in revenue", and then I look at Unity "Pay us $40 per month in order to have a dark mode and remove the splash screen," and I'm kinda annoyed a little bit.

 

I just wanted to point out, before going out, that 'rigging a high poly Zbrush model' isn't the problem as much as me having doubts their tech can also reduce the computational cost of animating all those vertices with that Nanite 'poly crunching' technique of theirs. Plus I doubt it can determined if bent joints look good or which detail can be extracted before animating a mesh. So it may just be tech for 'hard-surface assets' and not animated meshes and possibly not even tree rendering.

But even then, having everything but characters operate and not be constrained by scene polycount is a huge time saver for production.

 

They are explicitly mentioning environment though, not character, I don't expect that to translate to character yet.

 

I'll also elaborate.

This UE5 demo to me instantly looks like "I can use this". I mean, I want the GI.

The ECS thing, however, makes me remember people who say that you should learn Haskell to reach enlightenment. (Extra fun when they also can't explain what enlightenment entails). Basically, code is written for people to read it, and tools are made for people to use them. If there's a messy experimental workflow which produce higher performance results, it, to me, feels like somebody is being fascinated with "revolutionary new approach" while forgetting about users. It also worth remembering that you don't ALWAYS need huge amount of agents, so ECS is already an edge case of sorts. That's the reason for skepticism.

 

They never said that the tech is applied to the character or that character is made from quixel assets, so I think the character is done traditionally.

 

Digital Foundry apparently have some more note from the dev

also not meshlet based? and could run in ps4 without the GI? what are the trade off ?

 

Looks good, but when was the last time you saw a tech demo which didn't over sell and eventually under deliver?

I'll actually be impressed if it turns out creating such a scene yourself doesn't require an unjustifiable amount of development effort.

 

yes unreal looks like a game changer for that video. billions of poly? push zbrush model in a game directly (even for prototyping). and leave unity aside, cant even find a proper realtime gi solution on github / asset store (since these areas are ahead of unity devs most of times, sometimes experimental but ahead).

these are lots of reliefs for artists. no normals, no retopo, no lightmap baking & probe placement.
also virtual texturing, those particle interactions... i bet they'll have motion matching too.

buuuut on the other hand, this was unreal demo 8 years ago.

and lol cant even find a game with those particles & lava today. soo it could be "a little" exaggeration. but they say it runs on ps5 which is slower than my 2 year old high end pc.

and unity shows and then shares what it has. megacity and even today they released Heretic character(well it runs 45fps lol). they show what they really have.

i dont know, im confused. and there is more, megascans, 1million$ limit, epic games' store push. basically they're creating epic ecosystem.

 

Don't forget that they are probably working on their own scripting language, so there will be no need for C#.

https://www.reddit.com/r/unrealengi..._seems_people_at_epic_are_considering_adding/
https://venturebeat.com/2019/01/23/...ing-tool-maker-agog-labs-for-unreal-engine-4/

 

This is hilarious, though. They had unreal script before, and then proclaimed that blueprints are the only way. Yet here we go again.

 

This one is probably better than actual demo video itself:
https://www.unrealengine.com/en-US/blog/a-first-look-at-unreal-engine-5

So it seems to be heavy streaming / SSD usage here.

Stuff to be dumped (as a result):
- Normal maps;
- Lightmaps;
- Heightmaps;
(Potentially no need for the tesselation?)

Less shader ops as a result. (Potential free tripple detail map?)

I think actual install size would be less. But it probably will be more hardware (SSD) demanding than prior one. (Unless polygon count kept sane).

This explains a lot.

Also, Epic Online is free. Quixel is free. License is basically free for indie.
This tech is potentially scalable to the mobile level with the polycount being sane enough.
Which means there's less and less points to stick with the Unity as time goes on.
(Not that I'm instanly switching lmao)

Yeah, now I see, it is indeed a game changer.

For a Unity that barely supports proper streaming across the board - its a kick in the nuts.
(Hi there, async scene loading in the editor!)

Still, its nowhere near close to us, pune mortals. 2021 is too far away to judge.
Maybe something will change or improve in the Unity department.

Or go out of the preview. And HDRP gets fixed their bloody light shadows on the particles. REEEE

 

hmm...

I like what was shown from Epic for the tech, the new revenue agreement, and access to the new services for match making, cross platform across different stores those are really nice features.

After reading this thread I see the problem is not that Epic has new technology, its that Unity is extremely fractured and really buggy at the moment (with new features).

My team just fought for the past 3 months with SRP, vfx and shader graph that we finally made a somewhat painful decision to switch development back to the builtin render, post processing, standard particles (the game is stated to release with 2019 LTS)

The reason was simple the last several patches for 2019.3 have been development breaking for us. We made the mistake of assuming that packages labeled as verified on a tech release meant beta status, that is not the case. Major API changes are happening on all the new packages and interoperability between different systems not being compatible across patch releases.

We estimated when these features along with ECS were shown off around 2017 that they would coalesce in and around the middle of 2020 but from the few dev post in the last few weeks we seem to be pushing out past to 2021 sometime.

We have heard from the Unity devs that they have heard us loud in clear about buggy release but I wonder when we see SRP breaking between 7.1.8 to 7.3 in patch releases. To add more insult the, one of the videos put out from unity the devs admitted they are not really updating their own demo projects as new versions of unity come out. Unity does believe this is a weak point on their end though, but believing it and doing something about are two different things.

Now don't get me wrong I really like Unity but changes are needed for outside devs to better understand the direction of Unity development going forward.

 

BTW interview of Tim Sweeney about unreal 5

Yeah they are going to port fortnite from unreal 4 to 5 to dog food the process ... and fortnite is now a media delivery platform ...