@yujiang.wang In 5.7 substrate is enabled by default. The vrTemplate map uses a Blendable gbuffer for substrate. After running gpuLighmass, the results don’t look good - almost as if there’s no bounce light? When changing that gbuffer to Adaptive in the project settings, all works nice again. Makes sense?
Just in case someone uses that template as a start for a project (I did) and runs into the same issue.
Thanks! It works again. I do see a visual difference but that would be as expected? See video. I can send you the project if its something related to gpuLM.
Thanks! It works again. I do see a visual difference but that would be as expected? See video. I can send you the project if its something related to gpuLM.
Before my fix non-Substrate materials when Substrate was enabled were compiled to empty materials. Based on how the path tracer works I made them use the old, non-Substrate path. However I’m fairly new to Substrate and don’t really know how much the non-Substrate path vs Substrate-adaptive (which converts your material to Substrate automatically) will differ. Maybe you can compare how the path tracer behaves and have some idea.
The path tracer result when using blend and adaptive is totally different (see pic). Maybe some of my materials aren’t supported (hdr emissive sky)?! I will use the adaptive gbuffer for now until I’ve switched everything to substrate.
Currently I am porting older environments into 5.6.1 with your GPU Lightmass patch, pretty much all of these environments were baked with GPU lightmass in older versions.
When we try to bake environments with landscape actors in them, we get an instant crash
I’ve tried
-Converting the landscapes to nanite
-Splitting up the landscapes into smaller landscapes
Neither fixed the problem, however, I was able to use the merge actor tool to convert the landscape and its material into a baked static mesh, this workaround allowed the environment to bake but this results in a loss in both mesh and texture detail on the landscape.
Increase the value of r.RayTracing.ResidentGeometryMemoryPoolSizeInMB. In the latest 5.7.0 GPULM won’t crash but the geometries over the pool limit will still be removed.
Hi guys, one question, does this GPU lightmass versione from our lord and savior @yujiang.wang work with the experimental WP static light support? I tried but i cannot seem to make it work but maybe i’m just doing it wrong.
Tip: Improve real-time preview responsiveness in editor
You can add the following virtual texturing related CVars and settings to your DefaultEngine.ini to improve GPULM’s VT-based real-time preview responsiveness:
Thanks for putting the effort into getting this working. It’s very much appreciated.
Is it normal for GPU lighmass to force virtual shadow maps to be active when the plugin is enabled? Is it even possible to use cascade shadow maps and stationary lights here? The option to set the shadow map method in the project settings becomes greyed out (disabled). I even tried “r.Shadow.Virtual.Enable=0” in the ini, but if set my directional light to stationary, sure enough it’s using expensive virtual shadow maps even though the details say it’s using cascade maps. Am I stuck with CPU lightmass if I need stationary lights? I’m using GPULM 5.6.0 BTW.
Is it normal for GPU lighmass to force virtual shadow maps to be active when the plugin is enabled?
No, GPULM does not force VSM. If it says Shadow Maps then it is. If you’re using Forward Rendering then VSM is forced disabled. And CSM can be expensive specifically on stationary lights due to so-called preshadows. The traditional solution for this is to enable dynamic shadow distance for near-field shadows on stationary lights and use Distance Field shadows above a certain distance.
the bottom line of the engine is still console performance. amd gpus. you can throw a 5090 at a problem, but that tech is not gonna perform well on a ps5 (pro). technical fact. hmm.