GPU baker amazing feature but we need proper Realtime GI solution. i have some experiment on unity and result awesome(prebaked realtime lightmaps with reflection probe support) .
but unity using clustering phase on CPU ( per object lightmap resolution max 150x150 ish). if we got this feature in unreal within processing GPU (and unlocked lightmap resolution size) this will be really amazing.
Thanks, I was trying to figure out what in GPU Lightmass benefits from a better CPU, and what benefits from a better GPU. HWMonitor tells me that both my Ryzen 2700X and 1080ti are 50% to 80% busy during “Processing Mappings” on my scenes. , on most scenes is a 2nd GPU more beneficial (and yes, I am waiting for RTX / 2080 series card support in UE4 before I buy anything else)?
Combing ram & vram is not some thing easy as “I have ram. I have vram. Combine!” - it requires techniques called **out of core texturing and geometry. **Without care the GPU raytracer can be even slower than CPU. AFAIK out of core geometry is the main selling point of Redshift renderer.
On the other hand UE4Editor doesn’t support out of core geometry either - what it does support is out of core texturing (called texture streaming in UE4, with reduced accuracy compared to an offline renderer) and instancing. With the next overhaul of GPULightmass it will start to support instancing and limited form of texture streaming, which means as long as UE4Editor can open your scene then it should be high likely that GPULightmass can bake your scene from command line. Command line is still required because if UE4Editor & GPULightmass is rendering at the same time they are storing the same data twice.
An interesting thing is Lightmass CPU doesn’t support instancing either, and I was having trouble testing the Redwood scene I showed earlier with 32G RAM - I had to set 128GB virtual memory on my SSD to make it work. With the new GPULightmass:
17:12:27: Prometheus: Scene commited with 121 unique meshes, 2340321 unique triangles and 1315670 unique vertices
17:12:27: Prometheus: Scene commited with 19530 instances, 252685086 instanced triangles and 201084200 instanced vertices
17:12:27: Prometheus: Instancing has reduced triangles by a factor of 108.0x
17:12:27: Prometheus: GPU BVH total size: 177.23MB
And vram usage peaks at 3.7GB. So maybe when it comes out you will be able to bake your large scene.
Can confirm this last bit. There are levels in Red Matter which we could not render with Lightmass with anything above medium quality settings having 32GB RAM. With GPU Lightmass however, the 1080Ti has no problem. Sometimes it would ran out of memory, however plugging the displayport of your screen to your motherboard (so that it uses your CPU’s integrated GPU) or a second graphics card fixed this. Allowing the 1080Ti to be used fully by GPU lightmass.
Any ETA on the GPULightmass overhaul? GPU Lightmass is so good that we won’t ship any games without it!
I living Turkey and graph card is expensive. Using gtx 650 ti 1gb and not enough vram. maybe using ram and I can use gpu lightmass but I guess useless. I have idea; my processor is fx 6300 gpu lightmass maybe slowing down but fx 6300 very slow. Gpu lightmass maybe more speed than fx6300 ???
This problem appeared in the installation of Nvidia Tesla K 10. If you remove Tesla, everything is fine, even at the highest resolutions of textures. What can be done to use Tesla in the calculation of lightmap?
Dumb question, when building light from the command line, how can “Lightmass Resolution Adjustment” be set to 200% or more? The menu for it is (only?) in the Editor, correct?
In a game where I have a stationary directional light so it illuminates static and movable objects how should i configure my sky light so it behaves well with both movable and static objects since I see it’s not supported the stationary, or should I leave it Stationary?
Hi, great tools faster lightmass calculation, but im having problem with my shadows caused some kind artifact. Can someone help me cant figure out whats wrong.