I have a very strong interest in the use of GPUs in quantitative finance, and am in search of algorithms/simulations/models that can have their runtime heavily reduced by GPU computation.

What models, or segments of models, do you think might be interesting to explore with regards to runtime and parallelization?

Some things that are interesting, but have been done (not exhaustive by any means):

  1. Accelerating Cholesky decomposition when dealing with two or more factor models (2 factor Hull-White, LMM, etc.),
  2. Parallel algorithms for solving Backward SDEs,
  3. Various machine learning applications (of course).

This similar thread from 2012 seems to be slightly antiquated, and has not kept up with the growth of general purpose GPU programming in the past decade.

Summary: What is an area of interest where GPU computing can significantly make a difference, and warrants further investigation?