This talk is from October 2018, and so much has changed in the GOAI/RAPIDS ecosystem that it’s comical to see how much has changed! Regardless, the high-level concepts of how OmniSci works and the concepts behind GPU dataframes (then: pygdf, now: cudf) remain the same, so watching this talk still has value if you are interested in an end-to-end GPU workflow.
With the release of pymapd 0.7 a few days ago, getting started with GPU data science is just a matter of having an NVIDIA GPU and OmniSci Core (OSS) and a quick conda command to set up your environment:
conda install -c conda-forge -c nvidia -c rapidsai -c numba -c defaults pymapd cudf python=3.6
So check out the video, grab the Jupyter Notebook from the pydatanyc2018 GitHub repo and get started with GPU-accelerated data science!