Скачать с ютуб Hardware-Accelerated Containerisation with the NVIDIA Container Toolkit в хорошем качестве

Hardware-Accelerated Containerisation with the NVIDIA Container Toolkit

My talk on the NVIDIA Container Toolkit for April 2024's instalment of Techy Tuesdays, hosted at Society1 in Preston, Lancashire, UK. Original abstract: The advent and adoption of containerisation technologies, such as Docker and the subsequent OCI specification, have revolutionised how we develop and deploy our software; we can consistently run our applications and services between environments regardless of the underlying host's operating system or processor architecture, and the adjacent tooling massively improves developer experience. These portability and isolation benefits have seen the introduction of the container primitive to the GPU, with NVIDIA providing its own toolkit and runtime for running containerised applications in a hardware-accelerated environment. In this talk, I'll introduce the NVIDIA Container Toolkit and its underlying runtime before delving into how it can be—and when and why it should be—used to run containerised CUDA applications such as TensorFlow models and Blender rendering pipelines, complementing existing Docker primitives and tooling. I'll also touch upon how it can be integrated with Kubernetes, allowing us to cluster our hardware-accelerated software across complex node topologies.