Upgrades! Here we have 5,000 Static Mesh Components simulating physics on the CPU, all active all the time. Perfect collisions. Nanite, Lumen, UE5.3.2 I found a nice bump going from Actors to Components for simulating mesh. Colors change on impact/jump due to low Z UnrealEngine

No limits on what these meshes can be, aka they can be concrete, bricks, cars, columbs, etc. As long as we manage the materials a little bit. Nanite

i love seeing you and @SebAaltonen push the limits of what's possible and sharing your findings

@SebAaltonen Indeed, I’m trying to share more daily of what I am up to, and maybe weekly do a longer blueprint run through for youhube 🤘

it's so interesting how the mental models proposed as "free" abstractions, really _DO_ have significant performance implications when you're trying to do things as immediately as possible

I was trying to use 500 cpu, this could raise that to 2000... which would be great!

how are thread pool sized on cpu? is it hard coded or % based? like, can a threadripper really chew through it more than a 4-core processor?

I’m using sync physics too, but I believe they break it up rather well across all cores. At least I’ve seen load across all the 12900kf on packaged builds. As much as I thought physics was just one thread.

The 1080ti is on a 8700 - both systems seems equally stressed with 5000, which would make me think there is a single core limit somewhere

let's play: spot the bottleneck!

Oh that’s the point of all this, it’s all been finding where physics tick is , in Unreal insights. Testing and looking at when and what calls are going on, and looking at alternatives

Unity has an amazing profiler:(same as insights) and I did two years in there non-stop. I think I just wanted to torcher myself, not using insight the last 3 years and just going off feel 😵

🤣🤣🤣

Ill take what I can get!

If your use case is not running at 60fps on a dev machine, you might as well throw it away. I've seen instances of hundreds of thousands, even touching million objects with physics running on Unity.

yeah, I started there. The unity 3d dots sample is 1 million spinning - NON COLLIDING - cubes. The key here, is correct collisions. We can totally take a GPU instance, spin it, and do 1 million of them. But they dont interact. These fully interact, thats not something you see.

I went a little nutty

some "people" ask me, why not 60fps. The real question is, where did it crash. When did it crash. Whats the limit. Seems the answer to that, is just lower fps. That IS a great answer.

You can push it further with skeletal meshes + physics bodies on each bone 😉

Did that back in novemeber, god I hope I make a game and dont die

bad video, it got much better than this... but finding my own content on twitter ... lol

Chaos bodies do not need to be actors or components if you suffer through making them directly: (very rough experimental setup) Rendering them is done through simple Niagara mesh arrays in here in another processor

!!! Wonderful point MegaFunk! Glad to meet you! I am doing all this in Blueprints, not C++, so I may not have access to that without making a custom call. Can you confirm?

Great info, now I have another level of worry (that I should be doing these as chaos bodies) :D

Have you tried using HISMs here?

Thank you Adrian! That is what I am doing, testing it out. I found when using Nanite, HISMs are not needed and slower than ISMs.

Good to know! Yeah that's odd, maybe the LOD switching with HISMs adds additional overhead 🤔

It does!