So I have seen game devs freaking out after seeing UE5 demo. Have heard things from beginners like: “After seeing this there is no doubt for me, what engine to start learn an practice on” and “Im switching to UE5, it's the Future” and so on. I do have to admit it's really impressive tech and it does look like the future of graphics in gaming. But in same time I don no like this monopoly. I challenge all other big engines (cryengine, unity, unigine, etc.) and open source devs to beat UE5 until its set release date “late-2021”. You can do it! And even better.

P.S. I would contribute to such project. But it's definitely not a one man thing.

Oh, I am definately against monopolys, but you are seeing it wrong. This is not creating a monopoly, but breaking it

Let me explain: If there is any game-engine that has an unhealthy monopoly, its Unity. I personally hate Unity, even after (or should I say: especially since) working with it profesionally for 2 years. But its really hard to get to work with anything that is not Unity. In austria where I am, its like 95% Unity all the way.

So in that regard, the Unreal-Demo is a good thing. Truthfully, I tried convincing my bosses to make an Unreal project for the whole two years, but it was not before that demo that we finally got into the “well, we could actually try that”-territory. And I'm happy for that.

So no, I would not like Unity to counter that. Maybe some of the other engines, sure, but I'm happy that Unreal has that very obvious “AAA-graphics” to make it stand out.

Juliean said:
I personally hate Unity, even after (or should I say: especially since) working with it profesionally for 2 years.

+1 just for this sentence ?

trsh said:
open source devs to beat UE5 until its set release date “late-2021”

This would not be possible thought. If Epic has just one strength then it is their financial position thanks to Fortnite and their experience in developing their Engine since end of the 90's. You would barely get a similar product “until late-2021” and keep it open source/ well maintained by similar professionals that make Unreal Engine. And at least, they're pointing out by themselves that they started development several years ago, so …

But anyways, feel free to join our open source Game Engine project @trsh , I'm expecing your PM ?

trsh said:
I challenge all other big engines (cryengine, unity, unigine, etc.) and open source devs to beat UE5 until its set release date “late-2021”

Probably not going to happen. Not because it's impossible, but because Unreal has a ton of money, the expertise, and different goals. Epic is not only targeting the game market, but also TV productions, movies and animation, architectural visualization, and more. It's in a transition from a game engine to a full DCC tool. As far as I know, only Unity has similar goals, but they are way behind and have to sort their own problems right now (finishing DOTS, their new graphics pipeline, and so on).

Also, while the demo is impressive, “late 2021” is somewhat far away. Many things can happen in the meantime. I don't think other commercial engines will match Unreal in terms of completeness as DCCs, but they will probably feature many of the things that Unreal has for game development. The main downside of other engines is that they are not as accessible (in terms of cost, mainly), but tools like Decima (Hozizon Zero Down), Luminous (Final Fantasy XV, Project Athia), RE Engine (Resident Evil VII e Village/RE 8) and Frostbite (BattleField V) are as capable as Unreal, if not more in some aspects.

As far as I know, the closest (for a generous definition of closest) open-source options are Blender (not an engine, totally missing a game/interactive mode right now) and Godot (really small in terms of money and manpower, but making great progress). I'm following Godot and the next version (Godot 4) is getting really interesting, so you might want to take a look at that.

trsh said:
I challenge all other big engines (cryengine, unity, unigine, etc.) and open source devs to beat UE5 until its set release date “late-2021”. You can do it! And even better.

Thanks for the spirit. The day all games run on the same engine is the end of my definition of gaming : )

trsh said:
I do have to admit it's really impressive tech and it does look like the future of graphics in gaming.

I think it's current leadership in shown progress. I expected to see this from some AAA custom engines without legacy bloat, but now i'm surprised it's Epic who truely moves on, while others just seem to scale up.

But assuming we can keep improving technology forever, even long beyond things like ‘mostly photorealistic graphics’, we can always ask what's next, to proof Epic is not the only future out there.

UE5 has some serious issues:

Geometry:

Finally showing a true lod solution is really nice, but hiding lod transitions at subpixel level is still no great solution because it might use more data and performance than necessary. We need to see how their stuff works with lower polygon density.

Making a whole engine technology based on the idea of instancing models just turns the detail limitation into a content limitation. Just duplicating the same 10 rock models over the whole terrain is not unlimited detail, and causes serious limitations on low frequency variation.

Increasing geometric detail means increasing production costs. Free Quixel won't solve this.

Lighting:

Finally showing infinite bounces is really nice, but lag of a second to update is not really realtime lighting to me.

Mixing bunch of hacks like dynamic objects being only occluders, detailed objects being restricted to static SDF, and doing still so much in screenspace to get some high frequency details is… well, i'm not impressed. (thankfully)

But beating anything that was shown using RTX really kicks ass. I do not really like the way some HW vendors present themselves as solving all our problems for us.

I hope UE5 is not the only surprise we see with next gen. So far Epic is the only company that managed to impress me in that sense. Pretty disappointed from first Sonys game reveals, for example.

( now it's enough rant ; )

I look forward to seeing more demos of that LOD system. Remember, they're running on a PS5 system with 16GB of RAM, accessible to the GPU, the I/O controller, and all 8 CPUs. Plus disk is all SSD. The question is how much they're precomputing and preloading. The hardware supports blasting precomputed vertex buffers from the SSD right into memory the GPU can access. Then they basically do GLOD in the GPU, right? There's nothing wrong with this. It's just a very memory intensive way to do LODs. The question is, was all that done by preloading? Or will this work moving around in a big world?

The end of the outdoor scene in the UE5 demo is interesting to look at very closely. The viewpoint moves so fast, and there's so much motion blur, that you don't see much detail. But at the end, she stands before the gate. Step through frame by frame and look at the columns and stonework to the left and right of the gate. Watch the level of detail increase as the LOD system catches up. It looks like the hard case is being handled.

I still think open source gods can beat UE5. Just because their minds work differently ?))))))))

Nagle said:
Step through frame by frame and look at the columns and stonework to the left and right of the gate.

Tried it, but could not see any switches :?

However, if you scroll through the debug view shown at the beginning, you can see clusters of triangles switching between discrete LODs. so there is no progressive mesh or something like that.

Nagle said:
The question is how much they're precomputing and preloading.

I assume, on top of the triangles they have a point hierarchy and render it in compute either with splatting or frustum tracing. So for distant objects they need to stream only the top levels of hierarchy.
But not sure about points - if they have triangles all the way up this would ease compatibility with lower power systems like mobiles that can't handle per pixel geometry details, and it might end up more practical in general aqnyways,

However, we see the scene is composed from only a small number of models: Few rocks, few towers, one vase, even the mountains in the background are all the same model duplicated 3 times, things are mirrored symmetrically, etc.
This ofc. helps a lot with caching and memory savings as well.

The real limitation seems storage space not speed. So i'm sure we will see repetive geometry everywhere in the near future, just as we see repetive texture now.
It reminds me on the task: You have to build a huge world from Lego, but you are allowed to choose only 10 different bricks, and build it all from that limited set :D
So in practice we might see more models and variation, but less insane detail than shown.

Some sniffers have found out they use some variation of this http://hhoppe.com/proj/gim/​ and https://pages.jh.edu/~dighamm/research/2004_01_sta.pdf

Nagle said:
they're running on a PS5 system with 16GB of RAM, accessible to the GPU, the I/O controller, and all 8 CPUs. Plus disk is all SSD. The question is how much they're precomputing and preloading. The hardware supports blasting precomputed vertex buffers from the SSD right into memory the GPU can access.

Will definitely be interesting, also for our own work, to read their first talks and blog posts on GDC 2022 and earlier. I hope PS5 still provides native fiber access and Unreal devs got to a solution to also implement them for all the other platforms so we have a real chance to see multithreading and real micro-task based approaches in action without any/much scheduling from the OS.

Swapping all these stuff in and out of memory should not be the real problem with our known options. At least memory mapped I/O should do the trick but I know from my old days of working on the PSP that there are some tricks to also speed things up on the console that you will never be able to on an OS driven PC or Phone.

trsh said:
I still think open source gods can beat UE5. Just because their minds work differently ?))))))))

If Epic really keeps it current route of UE4 then chances are high that we can access it's source code from GitHub as well as we can with UE4 already. But anyways, the overall issue still remains, Epic has the money, Epic has the pros that earn the money to do their jobs right and finally they have at least 2 years any new development from another team needs to catch up.

Open Source devs mostly are doing this as a hobby or just because the product is so trivial that they don't need to fear someone can take their work they pay for in development hours and money that nobody else will really make big business wih that. So they lack of time/money because even a Student needs to get some food and roof to sleep well in the night so it'll take 10 times longer to do the work in a hobby project than in a regularly payed job and finally even if open-source has a huge community, it isn't ensured that you'll get a team of avarge size to get everything done and/or quality may suffer from contributions. A real company can catch all of these up by offering positions that specialize in different topics while such an ambitous project often needs all-rounder guys not run in dry mode.

Speeking from experience, our 4 peoples team already has a good amount of management overhead.

And even if a compan, one of the big players, decides to get into the engine market, they won't offer their product as open source solution I guess rather than something similar to the Unity model and then you won nothing at all ¯\_(ツ)_/¯

So still, if you believe putting the effort into a quality open source product is worth it, as I do for our work but from a different motivation point, then feel free to visit our Discord and join our team and bring your expertise in …