It’s hard to find concrete data on how RT cores work, but it feels like the architecture responsible for accelerating Bounding Volume Hierarchy (BVH) transversal might likely also be a natural fit for ray marching acceleration? Raymarching loops through an undefined number of steps running the same test at each stage until it finds the hit point, which is fundamentally a branch. My understanding is this is also basically the way BVH transversal works…
Is there enough information out there about RT cores to know whether I shouldn’t be getting my hopes up? We’d need software defined test location specification and software defined hit function. Is there a chance that will ever be possible on RT cores?