This was my first thought too, however when checking the registry for my TDR delay time, I found it’s already set high at 120 (decimal) which I believe is 2 mins? The strange thing is, when I delete the landscape and try baking everything is fine.
Also worth noting that this scene does not timeout/crash in previous version of Unreal with GPU Lightmass.
hi @brisck1 ,
We all wrote about this and in the early years of LGPU we went for a minimum of 300 if you had an early GTX 970 you will need over 500 be 600. My graphics card is Nvidia GTX 1050 and left at 620
MS does not recognise the problem, but drivers know it exists.
Just business as usual
Does UE5.4 have a raytracing light cap that also applies to GPULM? And is that cap changed or lifted in 5.5? Having some lights not bake in my test scene in 5.4, but they do in 5.5.
However, I have broken UVs on some objects in 5.5 and I can’t seem to resolve those. Still doing some testing.
You are a star… I had three meshes in a massive level (15k+ objects) which had a Material set to None… Ended up running a debug build where I could see the mesh that was causing the issue but your “hint” was the key!
Are there any plans to incorporate spherical harmonics for the volumetric lightmap into GPULM?
If not–how can I add it myself?
My first guess is to modify LightmapPathTracing.usf, but honestly I don’t have any idea where to look. There are a lot of different variables here and I’m not sure where to look.
I am using Engine Version 4.27 and I am using the version of GPULM that comes readily with the engine.
They still haven’t fixed the error: Assertion failed: GeometrySegmentIndex == RayTracingMeshCommand.GeometrySegmentIndex in version 5.5. It occurs every time I try to run the lighting build.
I had the same problem and also fixed by checking if any of the meshes in the scene had an empty material slot and assigned something to it. also unchecked the share material code that sometimes crashes packed games using dx12 and raytracing
I never used a so buggy software as the Unreal Engine. I render the same scene with the same settings at the next day and the GPU Lightmass crahses with the error:
hi All, @Matthew.Ivey ,
There is a problem which will impact
For users on NVIDIA 50-series GPUs experiencing issues after updating to the latest drivers.
Anyone tested GPUlm in 5.6.p1 ? I’m getting constant GPUlm crashes and if it doesn’t crash, the result looks bad. Testing using converted and fine-working 5.4 and 5.5 projects. Also, instanced meshes (in a BP actor) render as completely black.
Any suggestions - besides waiting for p2 and crossing fingers?
I’ve been getting erroneous lightmap output for all ISMs for at least the last 3 versions (including 5.6) of the editor and I’m still running into the bug where GPULM sometimes doesn’t generate a volumetric lightmap.
Doesn’t look like any of the outstanding problems were fixed at all.
Despite the earlier promise that Epic wasn’t abandoning GPULM… the evidence seems to be to the contrary.
There are major changes of code for 5.5 , could not make a Luoshuang’s GPULightmass version for LGPU 5.5.
The code seems to have settled down, many parts of LGPU are now in 5.6 which had been removed in 5.5.
The only thing that The Ray Tracing Final Gather Shader file seems removed in 5.6, but not needed for LGPU
The code used in CPU/LGPU is still the same.
hi @motorsep ,
So here is the relevant patch since the 5.6 previews, there have been 5 patches since 12th May relevant to rendering, because I keep track of LGPU 5.6, which is currently being built
This file changed on 5.6.0 for the State of Unreal 2025
Hi, after a long away from keyboard due to some health issues I’m finally back, though I’m not at Epic anymore. The quality of the built-in GPU Lightmass has noticably degraded since 5.0 as I was the only person working on it (while having health issues). Now, while I’m not at Epic, I can however still work on patches for the built-in GPU Lightmass as a hobby project. Like how the OG Luoshuang’s GPULightmass (LGPULM), this version installs by dropping files into your engine installation’s Engine folder.
Oh and one thing, you may ask why I’m not going to work on LGPULM, while it is beloved by the ArchViz community. Well, due to its architecture, it is very hard to add new features (like accurate materials instead of a UV0-1 square bitmap export of your material) or fixing existing issues (especially, out of GPU memory because of no support of instancing and agressive radiosity caching. Both techniques work very well in small scenes which is probably why it’s popular for ArchViz). So, basically I’m saying the built-in GPU Lightmass (EGPULM) has more potential - and one more reason is currently it is not properly leveraging the power of hardware ray tracing. Benchmark of raw ray traversal performace shows it should have performed as good as LGPULM even without those complicated techniques, which isn’t the case now, probably because my implementation is like, crap :). Let’s see how it goes.
P.S.: There’s a bug in 5.6’s ray tracing CPU instance transform upload code. If you find your foliage generates very dense shadows set r.RayTracing.InstanceBuffer.RLE 0 before bake.