Support for the Nvidia GPU-powered virtual machines will be available in three Google Cloud Platform regions -- us-east1, asia-east1, and europe-west1 -- with support for creating GPU virtual machines using the Cloud Console being coming next week.
Users can now attach up to eight GPUs (four Nvidia Tesla K80 boards) to any custom Google Compute Engine virtual machine. Each GPU has 2,496 stream processors and 12 GB of GDDR5 memory.
These GPUs can be used to accelerate a variety of computing and analysis workloads, from basic tasks such as video and image transcoding, to more complicated processes such as seismic analysis, molecular modeling, finance, fluid dynamics, and visualization.
The GPUs will support a variety of machine learning and deep learning frameworks such as TensorFlow, Theano, Torch, MXNet, and Caffe, as well as Nvidia's own CUDA platform.
Pricing is competitive, with each K80 GPU attached to a VM costing only $0.70 per hour per GPU in the US, and $0.77 per hour per GPU in Asia and Europe. Users only pay for what they use, with no investment outlays for hardware.