I am trying to create a benchmark demonstration for my university thesis.
The base of the benchmark is an image generating algorithm with ray-pathing implementation.
For now it only has spheres in it, as those are the easiest to check for intersections.
The three dimensional data is stored in a custom vec3 class, for which the intersection calculation is a small function that does a quick vector scalar multiplication.
I have seen that the new 30 series cards (Ampere) with 3rd gen Tensor Cores support TF32 precision format and data manipulation.
With that in mind, I want to use the tensor core to possibly speed up the intersection calculation in a single clock with tensor multiplication.
I was trying to find the details to use TF32 data format and the operation for it, but I was unable to find them.
Could someone help me with these informations as to where can I find them?
Thanks in advance,