I don’t think this belongs in here but more to the nVidia Developer forum. I already posted it there but that place seems like a friggin’ graveyard. Threads apparently go there to die. I also wasn’t sure if I should post this here or in content creation.
In any case… I’ve been working on a game with UE for a couple weeks now. A friend of mine creates the 3D models (or meshes). The game will feature destructible environments and I decided to go with PhysX/APEX to create the destructibles. Problem is… The standalone tool (PhysX Lab) uses only a single thread when fracturing the mesh. So it’s really, really, really slow when I use multiple depths (3-4). And the memory usage is quite high.
Is the stand-alone tool for content creation also single-threaded like the runtime?
Would I get better performance if I had an nVidia card, i.e. would the tool be able to use CUDA for processing power?
Unrelated to this: Should I use different versions of the SDK’s so that they are compatible with UE? I used the tool on a simple mesh and fractured it with two depths. It seemed to work fine in UE.
When using more depths the destructible will be more expensive to use.
To answer your questions:
I’m not sure if the PhysXLabs standalone tool is single threaded or not. Honestly, it shouldn’t really matter if it is or not since this is just a creation tool and not where you’ll use the DM that is created.
You would not see better performance from a NV card using DMs in UE4. The editor is card agnostic with this type of implementation. We don’t want to limit any part of the community with this type of feature out of the box by requiring people to have a specific GPU to use destructibles or cloth. So with that in mind the GPU really doesn’t offer as much a boost to performance with DMs as a good CPU with 4 cores or more would.
I’ve written a Troubleshooting and Tips guide over on our Wiki that may also be helpful as well. It covers a lot of the more common questions people have when they start using DMs.
UE4 works just fine, it’s the PhysX Lab fracturing tool that’s painfully slow as it only uses a single CPU thread and at least with my AMD card the GPU isn’t used. So with three depth levels and lots of different variations on the depth levels fracturing takes hours for a single mesh.
I was wondering if a NV card would use e.g. CUDA for fracturing and that way speed up the process?
I just found this new project that I really hope will be integrated into UE. Built-in PhysX Lab fracturing tool inside UE would be awesome!
Not sure on the CUDA part with PhysX Labs. Although I’ve never had meshes take hours to fracture either. If I were doing something more complex I’ve had it freeze for maybe 5 minutes or so, but that’s about it.
As for the user who integrated that, there are no plans to integrate this into UE4 without some clean up. It’s definitely a good start to getting PhysXLabs integrated, albeit with its own issues that would need to be resolved though. It’s something cool that I would love to see fully integrated in UE4 someday though! You can download and build the source for it here: https://github.com/EpicGames/UnrealEngine/pull/1113