Unreal Engine 5, [UE5 Developer Availability 2022-04-05]

[Remij]

Some developers will implement their own custom framework which is usually based on ECS to bypass Unreal Engine's game framework. Improving the performance of the default game framework would require a change to it's architecture which would involve a substantial rewrite of the engine and Epic has not explicitly revealed that they plan on doing this either in the future. The only thing you can do is to avoid the default game framework as much as possible since it's not very fast at updating the game logic or the actors. I have a feeling that Unreal Engine being this tightly knit to the OO model won't be sustainable for very long and it'll make it harder in the future to support multi-threading on the engine if they don't respond to the issue sooner ...

It's amazing how much better Days Gone on PC performs compared to the typical UE4 game, considering the quality of the visuals.

Bend and The Coalition's games all perform extremely well... and of course those are the two studios who use their own highly customized versions of UE4.

Unreal needs work under the hood for sure. Was hoping that UE5 would change things, but it's looking like any real changes are far out yet.

 

[PSman1700]

https://www.gamingbible.co.uk/news/...y50Ap3abLkJuVUMf0npuQyn4Dwq1IVBHMJtKWwhZaZGVo

'BioShock 4' Is Using Unreal Engine 5, According To Job Listing

 

[Clukos]

AAA using UE most of the time means "take the bits that work for the project and re-write the rest". UE5 looks like UE4 on steroids, it allows for faster iteration with nanite and lumen.

I think it got overhyped by the first demo they've shown and peoples expectations were off mark (understandably).

 

[SlmDnk]

Downloadable Lumen demo from the co-founder and technical art director of Wolcen Studio.

https://twitter.com/x/status/1399013729082261505

 

[Karamazov]

just a small corridor, not impressive.

 

[jlippo]

A Macro view of Nanite.
http://www.elopezr.com/a-macro-view-of-nanite/

A renderdoc view analysis of Nanite.

 

[cwjs]

AAA using UE most of the time means "take the bits that work for the project and re-write the rest". UE5 looks like UE4 on steroids, it allows for faster iteration with nanite and lumen.

I think it got overhyped by the first demo they've shown and peoples expectations were off mark (understandably).

What do you mean? This demo is much more impressive than the first one.

 

[Clukos]

What do you mean? This demo is much more impressive than the first one.

I mean about the engine as a whole. It's still very much Unreal Engine.

 

[DegustatoR]

A Macro view of Nanite.
http://www.elopezr.com/a-macro-view-of-nanite/

A renderdoc view analysis of Nanite.

This bit is interesting:

The information above also gives us an insight into cluster size (384 vertices, i.e. 128 triangles), a suspicious multiple of 32 and 64 that is generally chosen to efficiently fill the wavefronts on a GPU. So 3333 clusters are rendered using the hardware, and the dispatch then takes care of the rest of the Nanite geometry. Each group is 128 threads, so my assumption is that each thread processes a triangle (as each cluster is 128 triangles). A whopping ~5 million triangles! These numbers tell us over 90% of the geometry is software rasterized, a confirmation of what Brian Karis said here.

 

[Dampf]

What do you mean? This demo is much more impressive than the first one.

Nah. The first demo was much more impressive, atleast to the average gamer. Especially concerning the streaming speed in the last section of the demo and the texture quality. Maybe it was really possible only on PS5 and later on DirectStorage enabled systems? When using the drone in the new demo, it stutters noticeably at high speeds. it would explain why the first demo is not available for download.

 

[Silent_Buddha]

the thing about humans is they always carry a bagage of "implicit samples" for lack of a better term.

If you give a human 5 example rocks and tell them to draw more, wether you like it or not, their whole baggage of what other rocks look like still influences ang biases his output, and it can't be avoided even if they try.

Whereas an AI is a complete blank slate aside from the specified training-set. That's both a flaw and a virtue depending of what you are after. For creating art that looks plausible for humans, that's a flaw, but for finding unexpected/unintuitive but highly-effective solutions to complex problems, that lack of baggage can be its greatest strengths.

Another way to think about it is that the human brain is constantly processing hundreds and thousands of samples of a multitude of objects ever second. Every time you move your head, for example, you're seeing a new perspective (or sample) of an object. Take 2 steps in any direction and you have generated new samples of existing objects. Turn your head and you're sampling new objects, tilt your head and you've generated more samples of those objects.

For an NN, you feed these as discrete samples with some variance. As humans we're constantly being fed an analog stream of samples.

I think it's incorrect to think of humans having only a small set of samples compare to what is fed into an NN. The opposite is the case, IMO. We're constantly being fed thousands upon thousands of samples.

This would be true even for a baby. In cases where babies (or children) are fed only a few samples it can take many repetitions until some "model" is formed in their brain for how this works. Take for example something as simple as a parent moving something in front of their face. At first they'll think that their parent ceased to exist and for some babies this can cause significant distress. It can take a fair amount of repetitions before the baby forms a "model" that represents that the object being occluded hasn't ceased to exist, but instead there is something in front of that object.

You can even see this with older children or adults with things like the severed thumb trick (a simple illusion) if they've never seen it before.

Now, the thing that is neat about the human brain and NN's can't quite replicate nearly as well is being able to create new "things" from all these millions of samples that we've accumulated from our environment. The human imagination... Which then leads to human intuition... And the formation of new ideas...

Although it's sometimes scary (and exciting!) what machines can "intuit" (basically what it can find in a pile of seemingly random data) through NN that humans have a difficult time grasping due to the amount of data that has to be processed.

Regards,
SB

 

[Clukos]

The raster path only supports primitive shaders on consoles from the looks of it, it's hard to know how much of an impact this has in practice though. Packing projects with SM6 support looks broken as well, it's always missing some files that block you from launching the project, but to be fair the option is tagged as "experimental".

 

[PSman1700]

What do you mean? This demo is much more impressive than the first one.

Its running well across multiple platforms (more impressive then last years on-rail demo), its not impressive anymore.

 

[OlegSH]

https://imgsli.com/NTU5NzU
Playing around with primary rays RT debug mode in UE5.
Pixel perfect shadows, mirror and glossy reflections, Nanite geometry complexity - no lods, each mesh is at least 1 million poligons with 100% proxies (barycentric in the screenshots above is to show the density), 8 ms in 1440p on 3090.
The part I love the most is that the RT debug mode just works, lol. I am working on a demo to demonstrate HW accelerated RT stuff in UE5, but currently HW Lumen is glitchy as hell, though, there are tons of improvements in comparison with SW.
In primary rays RT debug mode, everything just works - reflections work as they should with lighting in hit points and there are shadows in reflections, shadows are pixel perfect without any aliasing.
Performance is 2x in comparison with HW Lumen + Nanite, so there is space for more things,

 

[Frenetic Pony]

Yeah, aside from Lumen and Nanite and the nicer UI, not much has changed which is a shame. I mean they are still using DX11... just why? Hopefully their DX12 implementation is much improved, it had so many issues in UE4. And no DX12 Ultimate support.

Speaking of AI, I was fully expecting AI accelerated workflow, stuff like ML upscaling textures and ML accelerated physics.

But it's early access so there's a good chance all the features I mentioned will be implemented in the final version.

Ultimately I'm satisfied from a user perspective, even if Lumen still needs a hell of a lot of work to be as relatively efficient as Nanite.

UE4 has been constantly updated and already has nice new features. As for NN physics, I'm severely cooled on that. Directly calculated physics are just... fast, and further totally accurate and predictable insofar as what you want out of them. Edge cases are already a bad nightmare, having to "debug" a notoriously black box neural net on top of that doesn't seem any fun at all. As for speed, like... offline rendered cloth levels in realtime envelopes level speed are possible, so who needs anything more?: https://web.cse.ohio-state.edu/~wang.3602/publications.html

I mean, gameplay and animation system overhauls, hypothetically good and efficient streaming, no more super low instance/object counts, multi user world editing, actual multicore support, this all stuff people using UE wanted and it's what's being delivered.

 

[Karamazov]

Exclusive battlefront 3 teaser

 

[Andrew Lauritzen]

I mean they are still using DX11... just why?

Please don't use DX11 in UE5, especially with Virtual Shadow Maps... as per the docs, performance will be very bad compared to DX12. It's there because it still works, but it is absolutely not recommended when using all the new stuff.

 

[cheapchips]

Exclusive battlefront 3 teaser

I'd expect next gen Frostbite to look at least as good for Battlefront 3.

 

[Clukos]

GPU Trace on a 3080 (at 4K native)

 

[Dampf]

Interesting UE5 talk here.

The performance target thing worries me A TON though. 1080p 30 with Hardware RT Lumen on Next Gen consoles. That's not acceptable, even with their new upsampler.

...Just why? I thought hardware-RT is faster than software RT at the same quality settings? Why not combine Medium Lumen + Hardware RT to get crazy good performance? It looks more than good enough. Or is triangle based Raytracing really that inefficient and their software solution is simply faster?

I can't wrap my head around this. THis is going to destroy performance on RT capable cards, instead of accelerating the effects like it should be, they crank up the settings to no end even though most people won't be able to tell the differences. I don't care if some effects are still in screen space. If I want 60 FPS on a low end RT card, I basically have to left its RT cores unused. Which is a shame. Atleast give us the option to use HW-acceleration to make Lumen run faster...