This talk will present our efforts in accelerating and scaling Empirical Dynamic Modeling (EDM), a nonlinear nonparametric time series analysis framework, using the latest high performance computing hardware and software. I will discuss our massively parallel implementation of EDM named mpEDM that leverages multi-threading and GPU offloading. mpEDM scales up to 511 compute nodes (2,044 NVIDIA V100 GPUs) on the ABCI supercomputer and achieves up to 1,530x speedup in large-scale causal inference computation compared to a previous EDM library. I will also present our recent work on a fully GPU offloaded implementation of EDM based on the Kokkos performance portability framework.