While I absolutely want to do that as well, the fixes have missed the 5.1.0 train and the earliest they can get into would be 5.1.1
3 Likes
Hi @YujiangW @Luoshuang !
I have made some very little tests:
Almost no changes between Irradiance Caching and First bounce options. Maybe in…reflections (?). In 5.0 changes were more evident (I don’t know if better or worse, but more noticeables):
Also, Indirect lighting intensity values don’t seem to work:
And if a Light is stationary, its direct lighting can’t pass translucent materials, but its indirect lighting can:
Thank you and best regards!
So IC and FB are methods to help with sampling. If your scene can be sampled pretty well without them you won’t see much improvement from enabling them. You can probably disable the denoiser and see if there is a difference in noise level.
Regarding stationary light shadows vs translucent materials, this feature hasn’t been implemented
Many thanks! I’ll give UE5-main a test in a few weeks and see if it resolved my problem.
Thank you @Luoshuang ,
I was just surprised because of the difference when using IC and FB in the previous UE version, where it was very, very noticeable.
And what about the Indirect lighting intensity value not working? A bug? A missing but future feature?
Thanks!
Just wondering if there is anything that could be done to get reasonable looking direct lighting using the current VLM probe system. I know it has not been designed for that, but there isn’t any other option right now if you are targeting lower end devices or high frame rates. As has been mentioned elsewhere in this topic, being able to configure the distance the highest density mipmap extends from a surface would be beneficial, as lowering the detail cell size often results in it terminating below arm or head height which is not particularly useful. Manually placing density volumes everywhere will fix the issue but that can get tedious. Edit: am I misunderstanding how the probe placement works? Is the first mip always a 4x4x4 block? What I am after is a way to increase density without lowering the detail cell size.
Also, and I may be showing my ignorance here - would it be possible to store direct lighting information as a lower order spherical harmonic, place them at a much higher density, then blend them with indirect probes? While shadow map level crispness is not necessary, getting the density high enough so that moving in and out of a shadowed area feels somewhat realistic and believable would be nice. The transitions are a little too soft to get that effect at the moment.
That’s kind of what stationary mobility is for. Of course it works differently, and you have overlap limitations, but as long as most of your scene is static, performance is quite reasonable. The shadowing of dynamic objects is what can be quite expensive (which can be disabled if the scene allows it).
I’m not sure what’s the issue you have with increasing density. A density of 50 is usually good enough for interiors with static lighting. Static lighting + capsule shadows if properly configured can be quite good (there are some useful cvars to change).
The rounded corners and strange penumbras of distance field shadows irritate me. Epic has done some work on improving the memory use/resolution of distance fields for UE5 which should help with the rounded corners, but I don’t know if that work has made its way over to the shadows, so I’ll have to do some more testing. The shadows produced by lightmass are of a very high quality and they have essentially no performance cost which is nice for mobile hardware and VR systems.
The problem with using a density setting of 50 (or lower) to get relatively crisp looking shadowing on dynamic objects is that mip0 ends below arm or head height, which is not ideal if your dynamic objects are mostly humans.
Edit: I understand regular stationary shadow maps are going to be the best choice in most cases, though there’s still a decent performance hit on lower end hardware like laptop iGPUs. I also like the large physically accurate penumbras that are possible with lightmaps, and there is no other way to get that look while remaining performant.
I think you’re confusing a few things. I’m not sure what you mean by mip0 here, and the volumetric lightmaps don’t do shadowing (except with capsule shadows).
With static lights, static objects get soft shadows. Movable get none.
With stationary lights, static objects can get either sharp shadows, or soft shadows (like the static ones) if you enable “use area shadows”. Movable objects can only get sharp shadows.
You could move your level a bit if the current volumetric lightmap samples placement is problematic.
This is just not true, volumetric lightmaps capture direct and indirect lighting, which includes shadows
What I mean is that for example a movable cube won’t cast a shadow into the ground because of volumetric lightmaps (without something like capsule shadows obviously).
Yeah, I think there is some confusion here about shadow casting and shadow receiving. Volumetric lightmaps do allow dynamic objects to receive some shadowing from static geometry (in a very soft, approximate way).
1 Like
TL;DR for people who wants to use GPULM in 5.1.0: Don’t.
Fixes incoming for 5.1.1.
For people who are willing to compile 5.1 from GitHub, they can apply the following source code patch file to get the fixes:
GPULM_fixes_5.1.0.patch (25.9 KB)
Theoretically the same patch can be applied to ue5-main, yet you probably need to delete the segments that don’t apply and merge them to the code manually.
3 Likes
TL;DR for people who wants to use GPULM in 5.1.0: Don’t.
You mean it doens’t work at all? Any specific issues?
Hi Luoshuang,
For people who are willing to compile 5.1 from GitHub, they can apply the following source code
I did this, but I don’t have much experience with compiling… Should I worry about these messages when applying your patch?:
$ git apply GPULM_fixes_5.1.0.patch
GPULM_fixes_5.1.0.patch:58: trailing whitespace.
GPULM_fixes_5.1.0.patch:67: trailing whitespace.
GPULM_fixes_5.1.0.patch:203: trailing whitespace.
GPULM_fixes_5.1.0.patch:278: trailing whitespace.
GPULM_fixes_5.1.0.patch:316: trailing whitespace.
warning: squelched 4 whitespace errors
warning: 9 lines add whitespace errors.
@Luoshuang tried to apply ur path. Sadly trailing spacer error and finally patch does not apply
Is the ue5-main already fixed? Can you post full file instead?
Edit, works on 5.1 branch! yay
@Dariusz1989 Just to understand: Did you have the same messages (trailing whitespace, see above) as me, and the patch still worked on 5.1 branch?
@Warner_V Yea the same warning.
I’m not sure if it worked 
I had so many other issues dint get to test. But GPUL did build. Give it a go if you want.
Is there a way to use the command line to build the lights with this plugin?
Have to agree - stick to the 5.1 preview if you can - dont upgrade to 5.1.0 if you need GPUlm. Lots of ‘broken’ lightmaps / strange colors.
1 Like