hello,
I'm actually working on a project, where i need to create texture for many different fbx assets, as it's a large amount of meshes, i've decided to create a structure robust and simple in terms of tweaking (switch from baked maps of Mesh A to B and so on). Things are i'm working with 4K Maps all the time, and it looks like SD can't get to work as expected ( Tremedous delay for the computing and crash often.
(my PC at home where i'sve setup that, is a last Xeon 8 Core Processor with 64 Gb Ram and a 1070 GTX. The config looks ok to work on 4K texturing, but i'm probably doing this the wrong way.)
So I've decided to create a new substance and create a folder for every bake maps related to their own fbx mesh.
Regrouped in folder MainBody_partA , B, C, and so on....

Then i've created one graph in the "custom node" folder to organise in a single node (0_BakeMaps_MainBody) all my different bake maps from all the different sets of assets. This way with an exposed Multi Switch (and Multi Switch Greyscale as well) input, i can select witch bake i want to render for the final texture this way.
in this exemple i got 4 maps on the left side related to mesh A B C & D, then Multiswitch node and then output.

On another graph (1_IDSelector), i've created a simple setup, with a material selector, and exposed the "pick color" to let me create the mask that i want in a simple and only one node in my Final Graph.

Then i've created the Final Graph "Mat_CEP02_MainBody" where i load my 0_BakeMaps_MainBody to link all my baked maps and also 1_IDSelector in the Blue Frame, to create on demand Mask for the related Material.
(At first I've instanced my 0_BakeMaps_MainBody Node multiple times all over the graph, and that look's ok , but for optimization, and to fix the crash that occurs a lot, i've decided to use only one node at the bottom)

Every output then are 4K as the main iput are 4K also. But at that time if i didn't suspend the engine, it crash at the computing of the Mat_CEP02_MainBody.
Of course for optimization, i've disable output(s) Computation in the Attributes of the custom nodes, because i only use them into the final graph.
At some point i've reach around 44 GB of ram usage for SD , until it crash..... It sounds weird. Is that wrong to do this way ?