Machine learning

CSC's computing infrastructure supports a wide variety of tools for machine learning and deep learning. In particular, compute nodes with dedicated GPU accelerator cards (Kepler K80 and Pascal P100) are available.

Taito and Taito-GPU

A variety of deep and machine learning frameworks can be used in the Taito super cluster, including the Taito-GPU partition. Note that, as with any other tasks, the Taito login nodes are not intended for heavy computing. Please use batch jobs instead. In particular, the Taito-GPU login node does not have any GPU cards installed.

Several machine learning related GPU-optimized Python packages (scikit-learn, TensorFlow, Theano, Keras, PyTorch, MXNet) are available in the Mlpython package collections. For more specific instructions on using TensorFlow, including how to use Intel's CPU-optimized versions at Taito, see the TensorFlow software page.

Caffe, Caffe2, and Dynet are also available as separate modules in Taito-GPU.

GPU flavors at cPouta and ePouta

The cloud services cPouta and ePouta contain several types (flavors) of virtual machines, including GPU accelerated ones with 1–4 GPUs attached.  Virtual machines are particularly suited for cases with specific requirements for the operating system or software, or when administrative rights are needed.  GPU-enabled Docker containers can also be run in GPU-flavor instances in cPouta and ePouta.