Game developer created a scene of his dog with 10 billion polygons in Unreal Engine 5

[jsilva]

What just happened? Unreal Engine 5 early access just launched and developers are already testing its limits, or at least trying to find them. Soon after release, a game developer loaded a 10 million polygon dog into the engine and replicated it 1,000 times, resulting in a 10 billion polygon scene that the engine had no issue handling.

To make this experiment, Gary Freeman, lead developer of Ionized Games, started by taking 122 photos of his dog sleeping on a bed. Then, he threw all the photos into Reality Capture, created 3D models out of them, and exported the generated assets with a 4K texture into Unreal Engine 5.

Once the 10 million polygon model was loaded into Unreal Engine 5, the developer used Nanite meshes to make 1000 instances of the model, creating a 10 billion polygon scene. As seen in the video, the engine did not struggle, running the scene at 60FPS at all times.

I was able to load a 10 million polygon photoscan of Ziggy in the @UnrealEngine. Using #UE5's Nanite meshes I was able to load 1000 instances of it at 60fps before I got bored. That's 10 billion polygons and it didn't even blink. It could have handled a lot more than this. pic.twitter.com/IMRnQIjFSx

— Sgt. Gary Freeman of the UNC (@IonizedGames) May 26, 2021

It's worth noting that Gary's system isn't exactly a toaster. He detailed his system features a Ryzen 7 5800X, an EVGA RTX 3070 XC, and 32GB of RAM. He plans to create a similar scene using a GTX 750, which he believes will be enough to handle it.

Despite the apparent complexity of the scene, the system wasn't close to full load. The developer also claimed that "any GTX 10 series card or newer is gonna see some mind blowing enhancements."

The developer was also asked about the asset's size, to which he responded it was 1.5GB. Gary added that games will "certainly get bigger," but the final size of a game is still the developer's responsibility. If he had spent some more time around the asset, he believes he could have reduced its size to 50-100MB without sacrificing its quality.

Permalink to story.

https://www.techspot.com/news/89846-game-developer-created-scene-dog-10-billion-polygons.html

 

[QuantumPhysics]

OK?

But what kind of GPU will you need to handle those graphics?

I'm thinking nothing less than 3090.

 

[Peter Farkas]

OK?

But what kind of GPU will you need to handle those graphics?

I'm thinking nothing less than 3090.

Does UE5 support DLSS?

 

[envirovore]

OK?

But what kind of GPU will you need to handle those graphics?

I'm thinking nothing less than 3090.

Know how I know you didn't RTFA?

 

[envirovore]

Does UE5 support DLSS?

Yes.

 

[Arbie]

But we still won't be able to Save Game. They forgot how to do that around 2008. I keep hoping they'll find that code on an old floppy.

 

[DrSuess]

OK?

But what kind of GPU will you need to handle those graphics?

I'm thinking nothing less than 3090.

In the fourth paragraph of the article it states

"It's worth noting that Gary's system isn't exactly a toaster. He detailed his system features a Ryzen 7 5800X, an EVGA RTX 3070 XC, and 32GB of RAM."

 

[RedGuard]

This reminds me of that Euclideon video:

 

[yukka]

If I remember correctly the new unreal engine does some very clever optimisation showing you only the surfaces that are visible to your perspective in a way that hasn’t been done before, allowing much greater complexity in scenes and the import and conversion of models based on real photos like this guys dog. Pretty impressive.

 

[kiwigraeme]

Will any male teenagers will be checking this out? the young mind is quite striving in it's own ways - obviously they will need to animate their creations

 

[Farkinell]

If I remember correctly the new unreal engine does some very clever optimisation showing you only the surfaces that are visible to your perspective in a way that hasn’t been done before, allowing much greater complexity in scenes and the import and conversion of models based on real photos like this guys dog. Pretty impressive.

Is this Mesh Shading?

 

[envirovore]

Is this Mesh Shading?

Was just going to ask if anyone knows if it's using Mesh Shading.

 

[Markoni35]

That's some fancy dog. My dog never had that many triangles. Not even when he was brand new.

 

[captaincranky]

It's worth noting that Gary's system isn't exactly a toaster. He detailed his system features a Ryzen 7 5800X, an EVGA RTX 3070 XC, and 32GB of RAM. He plans to create a similar scene using a GTX 750, which he believes will be enough to handle it.

OK?

But what kind of GPU will you need to handle those graphics?

I'm thinking nothing less than 3090.

You should probably read the article before posting crap about "needing at least a 3090".
Most active members are already well aware you have two 3090s.

So, why not post something like, "I have two 3090s, and my system could do that standing on its head", and be done with it..

Nice try at subtlety though, albeit a big fail.

 

[yukka]

Was just going to ask if anyone knows if it's using Mesh Shading.

I was assuming it must if it’s showing that much detail

 

[envirovore]

I was assuming it must if it’s showing that much detail

Not so much the detail aspect of it, but the performance that makes me wonder.
Mesh shading provides a massive performance increase.

Can't recall if the Digital Foundry analysis of the tech demo said it's been fully implemented and supported yet or not

 

[neeyik]

Epic's Nanite Virtualized Geometry doesn't require Direct3D Mesh Shading, but that's not to say the demos shown so far aren't using it.

 

[Arbie]

You should probably read the article before posting crap about "needing at least a 3090".
Most active members are already well aware you have two 3090s.

So, why not post something like, "I have two 3090s, and my system could do that standing on its head", and be done with it..

Nice try at subtlety though, albeit a big fail.

Still just two? I thought we'd be hearing about four. Shortages...