Clean and validate your cach in the swarm agent, and if it doesnt fix it, you may have reached your GPU’s VRAM limit. This error usually pops up (on my machine) when swarm unloads everything to your VRAM (easily checkeable with gpuz or even the performance tab of windows task manager) and it doesnt have enough room to fit all…
I’d suggest to lower some of your lightmap resolutions and give it a go again. You can check the total lightmap memory usage within the statistics window.
I just figured out what happened, for some reason all my meshes changed to cast shadows off and for some reason you need to have at least one mesh in your scene which is casting shadows otherwise GPU lightmass crashes. Took me like 2 hours to figure it out… but now it’s working again.
You have the **Packed Light and Shadow Map Texture Size **property in World Settings and your models’ lightmap resolution. Unreal packs together the ones that fit in the packed size.
Let’s say you set the packed size to 1024 (default) and you have a mesh with lightmap resolution of 512. 4 of these would fit in one lightmass texture. There are extra textures for the quality changing stuff and shadowmaps, but if you want to reduce the amount of textures, you either pack more models in less textures by increasing the packed texture size or you reduce your models’ lightmap resolution.
If you increase the pack size, it’ll reduce the number of textures in the end, but each texture is 4 times bigger, so if you’re also concerned about their size, it doesn’t make much difference. The biggest impact is to reduce models’ lightmap resolution.
About the packed texture size, remember that when Unreal needs to load the lightmap for an object, it needs to load the whole lightmass texture that model is in. So the bigger your packed texture size, the longer it takes to load. On the other hand, it has to load lightmap textures less often because more models’ fit in the already loaded ones. Also, a bigger packed size means you can have models with higher lightmap resolution (up to the packed size).
For huge environments with many meshes I usually go with 2048 packed size and try to never have a single model using more than 1024, unless it really needs it. And I try to break up things as much as possible so they don’t need big lightmaps to look good. Like, in a room, each wall is a separate mesh and floor and ceiling are individual as well. That way I could use 1024 for each. If everything was a single mesh, 2048 wouldn’t suffice.
And many times it looks like you need high resolution but you can get away with really low ones, like 32 or even 16 if the mesh is not too close to anything and has no detailed shadows etc.
Just remember to always keep your model’s asset lightmap UVs compatible with your resolution. If they are too tight, a texture size of 16 would cause light bleeding from one face to the other.
Hey everyone - I’m late to the party as I was on a mac previously. I just completed my first render with this amazing tool.
First is GPU, second is CPU. There are quite a few differences, including color temp. Also time savings. GPU took a third of the time it took to render CPU.
hi,
same issue for me. I thought the problem came from the mesh itself or from the way i unwrapped uvs, so I spent half of the day to figure it out, trying several methods of modelling and unwrapping. but no result. So I focused on my shader (a basic white material). I simply add a constant 1 to the roughness … problem solved. oO
There’s no such thing as more atmospheric. There’s just bias. If one looks significantly different from the other in terms of light transport (brightness, amount of color bleeding, darkness of indirect shadows), then the one or the other is doing something really wrong
Hi! I ask you if when the version 4.22 revolution arrives the lightmass calculation will still be CPU or will also provide us a GPU calculation. Currently all my lightmass calculations are in GPU and since I have to change my computer I wanted to know whether to invest in a CPU with a very high number of threads (36 in case the GPU calculation is excluded in 4.22) or medium (20 for the calculation only of distance fields). Thanks!