GOOGLE AND NVIDIA have made good on their promise to bring high-end GPU support to Google's Cloud Platform, opening it up for hardcore workloads like machine learning.
Nvidia Tesla K80 GPUs, already in use by IBM to accelerate its Power PC framework, will now be available for hooking up in an array of up to eight to power through crunching numbers using any Compute Engine.
Use cases suggested by Google include video and image transcoding, seismic analysis, molecular modelling, genomics, computational finance, simulations, high-performance data analysis, computational chemistry, finance, fluid dynamics and visualisation.
Available from three of Google's data centres, us-east1, asia-east1 and europe-west1, the K80s have 2496 stream processors each, with 12GB of GDDR5 RAM.
The service is aimed at users that need to spin up clusters on demand to run machine learning products like its own TensorFlow, but don't want the huge expense associated with hardware, which costs several thousand dollars per unit. Other supported frameworks include TensorFlow, Theano, Torch, MXNet, Caffe and NVIDIA's own CUDA.
This way, thousands of projects will be able to benefit from the technology and while it's not exactly cheap - with an hourly rate of $0.70 in the US and $0.77 in Europe and Asia, it's still vastly more affordable and accessible, billed as it is on a pay as you go basis.
Google adds: "Now, instead of taking several days to train an image classifier on a large image dataset on a single machine, you can run distributed training with multiple GPU workers on Cloud ML, dramatically shorten your development cycle and iterate quickly on the model."
The announcement acts as a curtain raiser for Google's Cloud NEXT conference which will offer opportunities to learn more about how to use the technology and integrate it with on-premise data centres, as well as cloud hosted ones. µ
Plus free apps from the Google Play Store
What we want to know is whether we can still get McDonalds delivered?
iSnuff movie was streamed live online
It's so close to the Apple Store, you'll be able to piggyback the wifi