GAME ENGINE DEVELOPER Reacts to UNREAL ENGINE 5 Demo

#UE5

[arcine]

Audio developer here: I'd like to chip in about the convolution reverb :).

The convolution reverb sounded great for all the reasons said in the video but there's a catch: it's ironically very static.

There are two common types of reverb: algorithmic and convolution.

Algorithmic reverb is relatively light on CPU, it's dynamic (the 'size' and frequency absorption of the environment can change over time) however, a good algorithmic reverb that can accurately simulate a real environment is very hard to create. They almost always rely on a complex network of delays (think of an audio echo effect), digital filters, modulation and feedback. Very fiddly stuff but powerful when it works. They can also be unstable due to the nature of feedback in the system.

Convolution reverb is often quite CPU intensive because it uses FFT (Fast Fourier Transform) techniques to efficiently convolve an input signal with an impulse response of a real environment (like a cave mentioned in the demo) and because of this, it's very hard to manipulate the impulse response in real time to make it dynamic. Convolution still requires a huge number of calculations on the audio samples because of the underlying DSP. You can interpolate between different convolution reverbs over time if the size or shape of the environment changes, but the problem with that is both impulses aren't 'aware' of each other and therefore don't interact how real sound waves would in that environment (resonances, phase cancellations at different frequencies, sound absorption etc). This is where algorithmic reverbs shine because everything in the network affects everything else in the signal path.

Unlike algorithmic reverb, convolution reverbs are very stable, predictable and they're INCREDIBLY realistic if the impulse was captured well. A good impulse response sounds/works great if the environment and character is static...these aren't. Getting a good impulse response of an environment is also very hard and involves a lot of expensive calibrated technology. When you do eventually capture an impulse response, you've captured that environment at that specific moment in time. That's it.

Fortunately most people don't care about stuff like this and convolution reverb works well in most cases. People will always prioritise visuals over sound for many understandable reasons, but you'll be surprised how a truly dynamic audio system can affect your experience in a game/film :).

[roarraizzer]

Finally, someone who really understands how things are done reacting to the demo.

[eggyrepublic]

I'm a 3D animator and I was like

Why am I even spending 10 hours ray tracing 5 seconds of footage.

[Kraxtn]

game devs "ten years producing these (shadows)"
every fps pc gamer "turns off shadows"

[travisbrown]

I'm a Game Design student and one of my professors recently revealed to us that he was on the team that did the animation for the character. Really interesting to get an insider's perspective about the work flow.

[abrahamciokler]

It's so refreshing to listen to someone with actual game engine knowledge talk about the possible algorithms/technology being used behind this demo. Unfortunately, most of the reaction videos out there are just journalist/YouTubers with no technically know-how and while they are entertaining to watch they are saying so much nonsense and just repeating what others say. Loved the video. You need to do a few more of these. Maybe like a series? Can I suggest you do a video of the ps5 specs and how the new hardware is going to impact real-time graphics/physics/sounds? Thanks.

[andygrng]

8:20 noo, can we get the one hour version? Or two, three.. I dont care, its so interesting

[larrypendleton]

I think the most impressive part of this demo was how well and detailed you explained everything.

[Sizzlik]

"I helped create the frostbite engine..and all i got was this lousy pen"

[ryanbarryaudio]

"Audio is incredibly important" Nice to hear some appreciation for us audio folks lol, thanks bunches!

[willwebster]

Watching him go through the flying bit frame by frame is nice as that's how I'm gonna experience this when it's on my pc

Edit: most likes I've had is about 5 so this is a new milestone, I'll be starting a Patreon soon. This comment thread is a nightmare enter at your owm risk

[DoomTobi]

Some of the details were already confirmed by some of Epic's graphics programmers on Twitter.

- There is compute shader software rasterization involved. Rendering such small triangles with hardware rasterization is slow, so they probably have a super optimized compute shader doing that. Maybe combined with visibility buffers instead of deferred rendering.
- They have virtual shadow maps with (roughly?) one texel per pixel that get also rendered with their micropolygon renderer. They use contact shadows/screen space raytracing to deal with the shadow mapping artifacts.
- Lumen uses a combination of screen space GI, signed distance fields and voxels.
- Also the demo is running with dynamic resolution scaling to 4k. It's rendered around 1440p most of the time

[グロ猫]

Cherno, yes, please upload your reactions and thoughts to the tecnical breakdown once it's out.


Also, could you make a video/series going through unreal engine 4's sourcecode and explain the rendering architecture and other interesting things?


Thanks for the Videos as always, very nice work!

[kazz]

I'm watching this like I have an idea what he is talking about.

[PabloNoriegaCoach]

hey man, unrelated to the video, but I just found your channel again after like 8-9 years because of youtube recommendations. your tutorial on creating a 2D java game helped me pass my first programming course in college all those years ago! good to see you're still doing videos. subscribed!

[adamz.]

Animator here: some thoughts about animation.

How high fidelity of uncompressed data will match the animation pipeline?

It's is great to see an improvement regarding animation in Unreal Engine 5. Contextual animation, motion warping and foot placement with IK is a good step forward.

Modern games character models (for example - Lara Croft from The Shadow of the Tomb Raider and Gerald The Witcher 3) have 70k to 120k triangles each. That amount of polygons is not easy to work with (in the context of rigging, skinning and animation.

Rigging and skinning (creating a skeleton that drives model and bounding it to mesh) is a pretty complex process on its own.
Zbrush can handle that many polygons while sculpting. Maya or 3ds max will collapse. Starting from Maya 2019 it's possible to turn on cached playback for animation. Another option is to get a model approximation (low poly) and then transfer animation to high poly mesh. However, all that is pretty complex and can cause undesirable results. Which leads to ending up with a lower quality product. All these issues may have a bigger impact if it's more that one character to animate in a scene.

The main character of that demo is stylised. It's clear from the opening scene. I will assume it's done to avoid the uncanny valley effect. Good example - Pixar or Disney movies.
In contrast, the environment in this demo is photorealistic. From ultra-high detailed models to photogrammetry.
That differentiation in styles makes her stand out from the main environment sometimes.

It will be interesting to see what approaches will be taken to adapt current standards in animation pipelines to match modern technology.

[tomhaswell]

I think I speak for everybody: I want you to talk about the technical breakdown :D

[flynnfittz]

Having someone with your knowledge on the matter to analyze this demo really brings out so much more information. This video was simply great!

[chrisbrent]

I remember ray tracing basic images back in the 90's on a pentium and for one basic picture it would render line by line and you'd have to go to sleep and wake up in the morning and sometimes it was still rendering.

[atub]

" i think they took all that Fortnite money and put it too good use"
I like how he said that soo much
😂😂😂😂😂👌👌👌👌👌