hey guys, I am currently working on a project that will probably require more than 11gb gpu memory to bake lightmass and wondering if there are any services that you know of
that might have higher end quadros and or titan rtx for rent online to do a good bake.
I am thinking to buy a new gpu (at the moment I have a evga rtx2080 non ti) it is a decent card but quickly realizing is not going to cut it for complicate scenes with 2k lightmaps, lot of textures etc for archviz. I was almost decided to get a Titan RTX but I’ve been following the news and seems like nvidia might have a new rtx3080 soon so I will wait a bit to know whats up with that.
In the mean time if someone knows of such service I will consider it in the near future, thanks for any info,
Jeez, how big are you making those lightmaps!? Try trimming it down, especially on props, you only need extremely high density on just a few areas, Ex: Flat, bare areas with direct, bright lighting, surfaces with relatively complex structure, even large ones under direct lighting like couches would be perfect even with 1024 or 512, the only way to tell is to give it a test run on preview, global illumination bounces set to 0 just to show direct lighting and speed up the bake. Set it to the lowest res that looks good, and due to you clearly going for high end, set textures with transparency (Including lightmaps if you see it fixes issues) to BC7 compression, its free texture quality.
Btw, shrink your textures to the point they look good but don’t lose detail, and see if some things look almost/exactly as good without some texture maps (Especially specular/AO, roughness too). Generic advice but even people working on major titles or ue4 marketplace dont do it. Its pretty much just a bunch of free vram and a bit of free fps.
This applies the most to flat surfaces (Doesnt need AO or specular almost ever) and grass (Which for some reason gets full pbr with normal map all at 2k/4k when you can literally just use diffuse at 512.)
hey Bits! thanks for answering yes I do agree with you on not making lightmaps unnecessarily big if not needed, optimization is always on my mind when I build a scene in unreal, that being using tileable textures when possible, map sizes as well as optimize geometry and lightmap size. I do come from film (clarisse, renderman…etc where we can throw almost limitless geo) so all this optimization has been a bit of a learning curve and still is.
I am doing some archviz work for for a client atm (an art gallery that will be presented in a 4k screen as well as VR) and definetly aiming for very high quality lightmaps (something like this I am thinking: https://www.youtube.com/watch?v=69v0TkBnvpA) some of the art pieces in the gallert will be seen from close, so I will need resolution hence higher processing.
atm already hitting between 6 to 8 gigs ram consumtion from the gpu and have mostly the gallery structure (optimized no back faces for walls for example) couple materials for test and varying size lightmaps (256 to 1k atm) and a ton of different lights, havent even try with 2k but I will most probably get out of memory with gpu lightmass as has happened before.
considering at the moment to buy a new card (titan rtx or rtx 5000 quadro) for this and future projects but a bit unsure because I will like to see what is nvidia bringing to the table with their new line of processors, but given the current global situation and their delay announcement…that might take a while. That is the reason I am potentially looking for a temporary solution to bake this project once is done.