Luoshuang's GPULightmass

Is the same bug present in the Epic GPULM (seen here: Epic's GPUlightmass - #481 by Arkiras ) also present in the 5.3 version of Luoshuang’s GPULM?

Hi Jumbohalo10, just a quick question, in ue5.4 your version there is not working quixel megascans? Or need to turn it on somewhere? Thank you

Hi @Olegsbr ,
The code for Unreal Quixel Bridge 5.4 plugin does NOT exist🤔
Even if it did i/you cannot distribute software not belonging to Unreal without Quixel permission.
Only the binary of the plugin is distributed.
A posting some years ago here showed how the hack the code for any version by changing a line.
I have signed an Quixel NDA, so I cannot do this.

The legitimate solution would be to download Megascan content using 5.3 Bridge into a 5.3 project then upgrade project to 5.4 copy over the Megascans plugins content to 5.4 from 5.3 which would then use the 5.4 Megascans plugins.

This would bypass the need for a non-existent 5.4 Quixel Bridge plugin😁

@Jimbohalo10, oh I see ))
Yes, for now I can add all the details to the 5.3 project and then open it in 5.4 version.
Anyway thanks for your reply!!! ))
OH, also I found that in 5.4 there is some issue with translucency, I attached two images wher you can see what I mean, so when you are close to the glass and looking through you see all the details of the shower fittings, but if you are moving back the details are loosing… I have no idea how to fix that.
Please have a look at the images, maybe you know this problem?



Hi, @Miguel1900 @maxbrown @3darchstuffs.com @Olegsbr etc all in this thread.
I am going to attempt building the 5.3.1 binary and find problems in translucent problems in 5.4
I am going to need a lot to test projects which I will store and test with, you can sen them via personal message if you want.

An example of really useful test projects is @Miguel1900
“BUG-Lumen_shadows_dissapear” initially for showing how badly broken 5.2 was. This was very useful in testing 5.3 and 5.4 and hence did NOT make a 5.2 version, but concentrated on 5.4.
Therefore I have decided that Screen dumps wont be any use. I cannot waste development time recreating a small project as a indie dev, you will have to send the zipped project and instructions for testing , I cannot investigate screen dump.

I am very keen on fixing problems with the project zip attached as I have a 90% record of fixing these
I very little interest and a 10% change fixing using in screen dumps only.
Thanks
Jimbo

Hi @Ozykz ,
I have check the code and its NOT affecting my LGPU 5.3.0 preview code. My version uses STRATA (OLD NAME) not Substrate code and would explain why 5.2 samples build fine in 5.3.0.

Its something for the 5.3.1 binary to look for.

My LGPU 5.4 binary version does not have the same code, and obviously that code cannot run Epic GPULlightmass.
Looking at 5.4 code its fixed in there as well when I checked the updates.

Hi @Jimbohalo10 ,

Test file already sent to you. Hope it helps.

1 Like

We are eagerly awaiting this LGPULM 5.3.1

Does this seem to mean UE5.3.1 LGPULM is not in yet.

hi @Miguel1900 ,
Well here is my 5.4 binary LGPU picture .Based on the earlier picture You can see that the top row of dots at the back are very much clearer.

However at the front bottom you have shadows on the back of the white balls. The version of 5.1.1 is more consistent with your original 4.26. I just wonder why there is an improvement on some bits of 5.4 and small shadows have disappeared

Using the High setting described in
Luoshuang’s GPULightmass (Changing LGPU Quality Settings)
namely

High
NumPrimaryGISamples=64
NumSecondaryGISamples=32

and

The console command was used to get Translucent working
The 5.4 missing white dot shadows are now back.

r.Shadow.TranslucentPerObject.ProjectEnabled

2 Likes

Miguel, this lighting is perfect! I believe it is using LGPU Lightmass in version 4.26 or 4.27 correct?

Jim, thank you very much for your commitment to keeping the original LGPU Lightmass updated, we are very happy and extremely grateful for that!

The question I ask is: why are these current results so different from the example posted by Miguel? the lighting in version 4 as he posted seems to have much better quality than now. Notice the soft shading on the right-hand wall… it seems like in all the other examples this has been lost. Not to mention visible global illumination in general.

Is there any way to achieve the same result in the next versions?

What has changed since @luoshuang’s source code update to fix large world coordinates (LWC) to work in Unreal version 5 and @pax_invictus’s frequent updates?

The first LGPU Lightmass updates in Unreal 5 looked identical to Unreal 4.

Once again, I thank all of you for your efforts to keep this working, as it is extremely important for us in the Archviz community, especially for those who are unable to use RTX graphics cards. Thank you very much!! :heart:

Hi @cvmabreu , thank you from my side. It’s a pleasure; At the end, I’m just another interested user.

Yep, that’s 4.26-27 (same result). But, ouch! I made a mistake (years without using that scene): the spheres were movable objects, to test volumetric lightmap, so those shadows were not baked, but from distance fields (read this, @Jimbohalo10 , and write me if you want the updated scene). I have updated the scene with movable spheres (the bigger ones), but also static ones (the smallests), to bake some diffuse shadowing overthere:

I have made some tests to also provide feedback for the Epic’s GPUL, so I have seen un UE5.0 GPULL was working fine, almost like in 4.27. However, in 5.1 there were artifacts, and it hasn’t been updated for 5.2 nor 5.3. I’m waiting for the classical ~30mb package for 5.4 by @Jimbohalo10

Got it friend. I’m also eagerly awaiting this classic binary pack but I believe it should only come out after the release of official version 5.4 by Epic Games.

Could you send me this project too so I can do some tests on previous versions (5.0 and 5.1) and also on the full early version 5.4 made available by @Jimbohalo10 on July 15th to try to understand the difference?

cvmabreu@gmail.com is my email.

This last image you sent was rendered in which version?

Thanks for the clarification

@Miguel1900

Hi @cvmabreu and all!

I have decided to release the project, so any one can do his tests and also contribute with the results and feedback:

PS: was baked in 4.26
PS2: @Jimbohalo10

Regards!

2 Likes

Thanks! I just downloaded it, I’ll do some tests soon and give you feedback.

1 Like

hi @Miguel1900,
This is the run of my LGPU 5.4 binary 17th July on the newly updated test example.
Using BaseLightmass.ini

;new values for HIGH discovered recently
;Lighting Quality High Scalability high
NumPrimaryGISamples=64
NumSecondaryGISamples=32

I have focused to the back to show the shadows on the back wall which is new for LGPU 5.4 . My hope is that when I have put the LGPU patches into the latest version (ue5main) I get at least this good a picture.

The screen dump below needs to be downloaded to actually see the shadows properly as its so high res.

What up, guys.

Did you guys know if Epic GPULightmass is broken somehow? I’m running a project with it and the results are below what I expected. Dark shadows and weird light. When I use the CPU Lightmass, works fine. By the way, any news about Luoshang Gpu lightmass for 5.3?

Thanks in advance.

Hi @Jimbohalo10 !

Well, it seems somehow accurate, isn’t it? Maybe just brighter.

Do you have binary files (few megas, to overwrite in the engine installation) of 5.4? So I can test it too in the latest Epic build, if you want.

hi @Miguel1900 ,
The 5.4 is same download LGPU 5.4 Local download
Ii used that to convert 4.26 to 5.4 LGPU binary.
The original 5.3 preview works just as well
all I changed at bottom of is BaseLightmass.ini

;High
NumPrimaryGISamples=64
NumSecondaryGISamples=32

I have been working on 5.4 (ue5main 17th October).
Git merge will not correctly copy over files, so looks like another 40 hours of hand coding. This will allow closer syncronisation when EPIC 5.4 BINARY comes out. Looks like end of November my version will be finished. Meanwhile I am doing all my testing on the Local 5.4 binary.