DXR is this DirectX Raytracing - Wikipedia
That means it’s DXR, not CUDA (which is what GPU Lightmass utilizes). Nvidia is pushing RTX further and Epic is on board with Nvidia. So I am sure that’s why it’s DXR and not CUDA (at least for now).
On older cards without hardware acceleration, both CUDA and DXR use compute shaders on the GPU, on cards with RT-hardware they can use both the RT cores and compute cores to do DXR work.
If the new GPU lightmass does a similar amount of rays and work, then it will be of a similar speed, your 1060 won’t suddenly become slower than cpu.
‘supported via drivers’ is the same as CUDA.
If anything it may be faster, as DXR is optimized for raytracing, while CUDA is more generic compute.
The fact that it’s DXR is good, as this means any gpu (ati, intel.) with DXR capabilities will be able to be used, provided they have the needed performance and Vram.
Them using DXR for it isn’t a ‘onboard with nvidia’ thing, it’s just the only practical thing to use with larger compatability. (Though one could argue that they should support the vulkan raytracing, but I don’t think that’s actually usable yet. Because DXR will be windows 10 only. )
It won’t ever go back to CUDA, because that means hard-excluding ati and intel hardware. (Extra so since ati is promising to show up with hardware raytracing themselves in the near future.)
However, the most interesting stuff to me is that the new one contains code for checking that meshes and lights and etc get changed during gameplay.
So this probably means runtime preview options in the editor. (And possibly, hopefully could mean runtime caching/baking ingame, but I kinda doubt that. Still hopeful though.)
One big thing this does mean however, is that RTX gpus will be significantly faster with the new GPU lightmass, as they’ll be able to leverage both the RT cores and the normal compute cores. While right now they’re just stuck with the compute cores only.
I don’t know about that @ I played Quake2RTX on my 1060 6Gb and I could only do so in 320x240 at 14 fps in lowest possible quality. So no, I don’t think DXR Lightmass will be faster than GPU Lightmass on my 1060 (and I think it will be slower than CPU).
Thanks, I actually know what DXR is, my point was that there is no information about the DXR lightmass and how they will implement it, also what kind of limitations it will have and so on. While GPU lightmass aka the one is tested and a number of peoples (me included) are quite comfortable using it. 
The code is on , actual implementation You can even build source and already mess with it.
This is cool to know, gotta try it sometime soon. Thanks for the info!
Playing a path-traced game in DXR is not the same thing as doing a bulk job like lightmapping.
Even in your example, your 1060 was doing 1 million rays, per rays/pixel. And was likely doing 5-10 million rays a second depending on what effects were being traced.
Compare that with the 100+ while doing a lightmass render.
The difference there is not CUDA vs DXR, but rather the massive performance difference of doing something realtime ( and needing to do hundreds of different things every frame.) vs being able to sit down and spray out hundreds of millions of rays to lightmap 1 texture, then going on to the next job.
The new Gpu lightmass might end up slower than the old one, because they’ll likely be rolling in features that were never implemented in the old gpu lightmass.
(Like soft shadows, IES, lightmass portals, stationary skylight, possibly/hopefully adaptive resampling. )
But on RTX cards it’ll likely end up significantly faster due to finally being able to tap into the RT cores as well.
There’s also some dynamic/runtime functionality in it that I’m curious about. Looks like we’re finally getting editor previews. (Or even possibly runtime-caching/baking.)
I’ll have to take a look at it sometime.
sadly this is not working for me
Hi guys, I have complied the next version 4.25 with the new GPUlightmass and I am afraid I can´t make it work on my 1070 graphic card. Not sure if the problem is due to how I have compiled it or that it´s not meant to run on non RTX cards. Below are screenshots of the errors.
Your error is that you haven’t compiled it, most likely.
Navigate to that path and verify that you actually have the UnrealLightmass.exe there, because that’s the error it’s giving. Failing to cache it can mean you just didn’t compile it.
I don’t think lightmass gets compiled by default if you just press build. You have to explicitly build it unless you add it to the include list.
Greetings community!
So, is last Lightmass 4.24.2 stable ? No bugs/errors ?
Hi Chris i have the same problem!! could you solve it?
I have a lot of problems with version 4.24.2 my bakes generate black spots and my skylight does not respond in static or stationary mode, I can not see any changes in the viewport and already reinstall the unreal.
with version 4.24.1 everything was going well for me! Does anyone have the same problem? Can somebody help me! thanks!!!
Whoa. As a plugin? Wonder when it will be available…
I guess we’ll get a preview at GDC 2020. Since they mention unrelease rendering technology preview.
Hey Jake,
No I have not found a solution other than working in 4.23 until it gets fixed unfortunately.
I have a problems with version 4.24.2
I can’t bake my scene with gpulightmass when I use a collab viewer template.
Thanks
Hey guys, I used this GPU lightmass with much success a few months ago, and now that I have reinstalled it to unreal (4.22, same as the previous time) I have noticed that its still fast but not as it used to be, maybe some new drivers are messing with it?
When I am baking I am using a very low % of the GPU (Yes I have checked the CUDA section) and now the same scene that took me 5 minutes to bake, now takes 23. Any ideas?
Ok, so can we leave beside this DXR scrapp and know how to downgrade to 4.24.1 for working GPU lightmass? Where can i get back 24.1?And please stop adding irrelevant comments like, how amazed you are, to this thread so we can se the important messages? I will delete mine afterwards.
very very happy
when we can download the preview of 4.25 ???
i Have big Probleme that i want to use my shared gpu memory becouse unreal give me out of memory message
i have 1070
up to date
install cuda Toolkit 10.2