I’m wondering, how one can define a second gpu via nvlink to render certain raytracing passes.
it should somehow work since 4.23 (https://www.unrealengine.com/en-US/b…-the-evolution, “the low-level support added in 4.23 offers control over which GPU will render a certain pass”)
we are really trying to push the graphical fidelity with raytracing, but cant get enough performance out of a single titan rtx, so i really would appreciate any replies
i know, that raytracing is an early beta integration, but if multigpu is listed as a new feature, one should find documentation on it.
Need out of the box support for multi GPU with access to the hardwares’ memory pooling functionality.
4 Titan RTX’s should be an option with 48GB of vRAM accessible without hand coding a bunch of stuff, but with the option to dictate which RTRT features are processed where. Maybe two features on one GPU, and another feature on one GPU, and another feature distributed across the remaining two GPU’s.
You’re going to have to wait if/when it comes
nope, not a pipe-dream. cause thats basically what ILM is doing in their custom branch now for some time.
Somebody send ILM a fruit basket or two … maybe they’ll champion this for integration in 4.26 (along with GPULightmass) !