We'll cover applications that range from the traditional HPC world to the more exploratory machine learning-based setup. On the traditional HPC front, we have tested the speedup and scalability of computational chemistry open source packages VASP and LAMMPS on HPC clusters fitted with GPGPUs and compared against CPU-only nodes. A performance enhancement of at least 3x is observed, which agrees with available literature. On the machine learning front, we're experimenting with Google's TensorFlow and the applicability of deep learning approaches for a set of challenging problems at Shell, including (a) searching for an optimal dispatch strategy under uncertainty, (b) searching for price prediction patterns in European energy data, and (c) fault detection in raw seismic data. The latter was presented at GTC last year. This time, we'll show scalability results using TensorFlow. GPGPUs have been key in speeding up model training for all these applications.