epic should fund this :]
Epic recently hired . There’s Lightmass developments being worked on.
Good news. İ hope Gpu lightmass support multibounce skylight and use RAM.
You can try what has made for 4.19 already, his “prototype” uses GPU VRAM and has multibounce for skylights. 
Not vram. Ram so system memory. Shared memory.
I’ve tried but can’t see multibounce support… You talk about 4.19.1??? I use 4.19.2.
Hi!! Actually it worked fine with 2 light scenarios…but now i have 3 light scenarios, and the third one is black…so it seems it doesnt work properly…and it probably just supports 2 light scenarios…maybe can give us some feedback ??? Thanks
Should be the same, this was baked with 4.19.2 and GPU lightmass with the settings visible on the bottom right of the image… And there is definetly bounces from the skylight (which is the only light source in the scene, and has an HDRI filled in).
https://i.imgur.com/lZUZZyD.png
Well seems that i spoke too soon… i build the light again and it worked for the 3 light scenarios…so its working
This is my log
10:23:42 AM: Preallocated 0.0Gb for kDOP nodes and triangles
10:23:42 AM: Building kDOP took 0.00 seconds.
10:23:42 AM: Static lighting kDOP: 6 nodes, 4 leaves, 16 triangles, 24 vertices
10:23:42 AM: Static lighting kDOP: 25.000% wasted space in leaves
10:23:42 AM: Preallocated 0.0Gb for kDOP nodes and triangles
10:23:42 AM: Building kDOP took 0.00 seconds.
10:23:42 AM: Static lighting kDOP: 6 nodes, 5 leaves, 16 triangles, 0 vertices
10:23:42 AM: Static lighting kDOP: 100.000% wasted space in leaves
10:23:42 AM: Processing…
10:23:42 AM: GPULightmass Kernel: Importing mesh: 24 vertices, 12 triangles
10:23:42 AM: GPULightmass Kernel: Embree SBVH Construction finished, 1MS
10:23:42 AM: GPULightmass Kernel: Convert to CudaBVH finished, 0MS
10:23:42 AM: GPULightmass Kernel: Line 75: cudaCheckError: unknown error
10:27:52 AM: [CloseConnection] Closing connection 067A48BF using handle 067A48BF
10:27:52 AM: [CloseConnection] Connection confirmed for disconnection 067A48BF
10:27:52 AM: [CloseConnection] Closing orphaned Job (3B80817F-4773214E-9607D697-ECB13F02)
10:27:52 AM: [Job] Job has failed! Job is closed while Tasks are still PENDING
10:27:52 AM: [CloseConnection] Orphaning local connection (067A48BF is the parent of 0931CFFC)
10:27:52 AM: [CloseConnection] Connection disconnected 067A48BF
10:27:53 AM: [MaintainConnections] Local connection has closed (067A48BF)
10:27:53 AM: [MaintainConnections] Removed connection 067A48BF
10:27:53 AM: [Maintain Jobs] Killing rogue Job “3B80817F-4773214E-9607D697-ECB13F02”
10:27:53 AM: [MaintainConnections] Detected dropped local connection, cleaning up (0931CFFC)
10:27:53 AM: [CloseConnection] Closing connection 0931CFFC using handle 0931CFFC
10:27:53 AM: [CloseConnection] Connection confirmed for disconnection 0931CFFC
Anyone help me? I use gtx 770 and I also change TdrDelay to 300 but it doesnt work
Maybe multibounce skylight is work but only vram work is useless. Because I have bad pc 16gb ram but very good gpu 11gb vram.
What would be a good card (or rig) to get for GPU lightmass baking that would be faster then a 1080? A titanXP? I need to decrease my build time for a project. Would CPU make any difference in this scenario? i7 vs Zenon?
Thanks,
Titan V or use swarm distributed rendering on multiple PCs.
I wonder if the V’s HBM2 memory would improve performance or not… Could be ?
Here’s a tip: if you have 2 or more GPUs, plug your monitor to the second GPU (and maybe restart your PC). This way, GPU 2 will be used for rendering Unreal Editor, Windows, etc.; and GPU 1 will be used solely for solving GPU Lightmass.
This way you can do other stuff with your display GPU while Lightmass works (like playing games, watching videos, etc.), and also greatly reduce the risk of GPU Lightmass running out of memory. I have checked and for a sample scene, GPU 2 (rendering) uses 3 GB of VRAM, and GPU 1 (Lightmass) uses 2 GB when building lighting.
Even if you don’t have 2 GPUs, it might be a good idea to plug your monitor into your onboard graphics, if you’re running out of memory when building lighting.
PS. Thanks Luoshang for this awesomeness.
How would you know wich one is GPU 1 and wich one is GPU 2?
It would be awesome to have a way to configure this on the GPU Lightmass module, looks it could be super useful.
Thanks!
the gpu in the highest pci e slot is gpu1
I check each GPU Usage in MSI Afterburner. But yeah topmost slot is always the first.
Hi, . I have a weird problem. When objects is too close for each other, after bake appears strange black dots. I noticed that if i will increase lightmap resolution it help.
But i was wondering is there any way to avoid this dots except increasing lightmap resolution. (I am using GPU Lightmass with Ultra High settings)
Hi!
It’s just lightmap res/uv problem and has nothing to do with 's lightmass!!
Yeah having a higher resolution lightmap will help! If you’re after results you shouldn’t “save up” on lightmap resolution!
Another workaround is to lay your lightmaps properly! Proper padding, pixel/grid snap… and you can always “cut off” the parts of the unused uvs to avoid pixelated/blocky contact shadows! …or snap the overlapping top object to the overlapped bottom object’s lightmap “grid”'s corners…
I got the same problem, however, the UV layout and padding are correct and the lightmap resolution is red… I don’t get that when using CPU bake, so it should have something to do with the GPU implementation.