MesosCon NA 2016

GPUs are the tool of choice for accelerating the data mining, machine learning and deep learning algorithms abundant in today's data-driven computing industry. These algorithms help solve problems ranging from weather forecasting to face and speech recognition to helping find a cure for cancer. In spite of this, Mesos has traditionally not supported GPUs as an allocatable resource. As such, many users have resorted to having a separate GPU cluster to run these algorithms outside of their primary Mesos installation. Fine-grained sharing of resources between these clusters has not been possible.

By adding first class support for GPUs in Mesos, developers are finally able to leverage the power of the GPU in the same shared cluster as all of their other workloads. This talk covers the details of how we added GPU support to Mesos as well as how developers and operators can now leverage this new capability.