Google introduces Deep Learning Containers

Google this week announced support for Deep Learning Containers in beta.

Containers have dramatically changed the way applications and workloads are deployed in recent years. The AI workloads are now no exception to this trend. Pre-built containers give AI researchers power by providing performance-engineered AI containers.

Installing deep learning frameworks is a complicated and time-consuming process. This limits AI researcher's ability to exploit new features from the frameworks. The complicated and time-consuming process of installing framework is now simplified with Deep Learning Containers.

Simple docker 'pull' and 'run' commands come in handy in using the latest applications. These deep learning containers are pre-packaged, performance-optimized, and compatibility-tested, says the announcement. These containers further provide a consistent environment while working with Cloud AI Platform Notebooks and Google Kubernetes Engine (GKE). The containerized environment makes it possible to scale in the cloud or shift across on-prem.

The announcement further stated the availability of hardware optimized versions of TensorFlow both suited for training on NVIDIA GPUs or deploying on Intel CPUs.

As of now, these deep learning containers only support machine learning frameworks like TensorFlow, PyTorch, and scikit learn. According to Google deep learning website, support for more deep learning framework will be available in the near future.

It is also worth mentioning the NVIDIA GPU Cloud (NGC) which also provides AI supercomputing containers. NGC's deep learning containers support a broad range of deep learning frameworks such as TensorFlow, PyTorch, MXNet, TensorRT, and more. Pre-integrated, GPU-accelerated containers include NVIDIA CUDA Toolkit, NVIDIA deep learning libraries, and an operating system. NVIDIA has tuned, tested, and certified these containers to run on select NVIDIA TITAN and NVIDIA Quadro GPUs, NVIDIA DGX Systems.