Google told its audience last year that it would be launching a support for high-end graphics processing units (GPUs) for machine learning and other specialized workloads in 2017. Today, the company delivered the word that it promised. On the Google Cloud Platform are the GPUs available for the developer community. These are NVIDIA Tesla K80 GPUs. With these the developer will be able to attach up to eight custom compute engine machines.
US-east1, Asia-east1 and Europe-west1 are the three Google data centers in which the GPU based virtual machines are available on. K80 core features are
- 2496 of NVIDIA’s stream processors
- 12 GB of GDDR5 memory
The K80 board features two cores and 24 GB of RAM. Complex simulations require more computing power and Google has kept this in mind while developing the GPUs. It can also work on deep learning frameworks like TensorFlow, Torch and MXNet of Caffee. The aiming audience for these new features is the developer community. People who like to spin up clusters of high-end machines in order to power their machine learning frameworks. These Google Cloud Platform GPUs are integrated with Google’s Cloud Machine Learning service along with various databases and storage platforms.
Google Cloud Platform aiming for developer community
The pricing by Google is a bit on the high-end but will surely set back mature developers a couple of thousand dollars. In the US the cost per GPU is $0.70 per hour and in the Europe and Asian data centers this price is $0.77. However, a Tesla K80 accelerator with two cores and a 24 GB RAM is a lot on the table in comparison to the price.
Google is expected to host the Cloud NEXT conference in San Francisco in a few weeks so this announcement is just in time. Some chances include hearing more about how Google will make its machine learning service available to more developers.
Image via Info World