Research

Ray Tracing is a popular image rendering algorithm used in graphics applications. Recent GPUs have specialized RT cores that accelerate sections of the ray tracing pipeline. Since RT cores are designed to accelerate ray tracing applications, it would seem that they are not very programmable. However, similar to how pre-GPGPU researchers modeled computations as shader programs to leverage acceleration offered by shader cores, I work on exposing the programmability of RT cores. 

We find that a wide variety of applications such as clustering, cosmology simulations, point clouds and other n-body problems can leverage RT cores to accelerate parts of their computation. We ultimately hope to create a CUDA-like environment to allow users to easily write programs without concerning themselves with what goes on under the hood. 

Check out my papers on this topic to know more about how all this works! 

Here's a hint: