Google announces Preemptible GPUs
Google has launched preemtible GPUs, a programmable logic chip made for display functions, which enable users to use virtual machines with GPUs at a 50% discount. Although, it should be noted the company can shut down the use of the service at any time with a 30-second window before the GPU workload is finished.
Google originally launched preemptible virtual machine instances in May 2015. The new feature enables users to attach Nvidia K80 and Nvidia P100 GPUs to preemptible virtual machines, which are currently offered on the the Google Cloud platform. The company will charge 22 cents per GPU hour for Nvidia K80 and 73 cents for Nvidia P100. Users must purchase a virtual machine as well. Additionally, users cannot use them for more than 24 hours.
“Preemptible GPU instances from GCP give us the best combination of affordable pricing, easy access and sufficient scalability,” said Woody Sherman, CSO, Silicon Therapeutics, in a statement. “In our drug discovery programs, cheaper computing means we can look at more molecules, thereby increasing our chances of finding promising drug candidates. Preemptible GPU instances have advantages over the other discounted cloud offerings we have explored, such as consistent pricing and transparent terms. This greatly improves our ability to plan large simulations, control costs and ensure we get the throughput needed to make decisions that impact our projects in a timely fashion”
Google’s offering mirrors Amazon Web Services’ Spot Instances offering, which gives developers access to virtual machines at a discounted price as well. Like Google’s offering, however, these virtual machine can be shut off without much warning.
According to Google, its GPUs are an ideal fit for machine learning and technical workloads that run in batches. “Preemptible GPUs will be a particularly good fit for large-scale machine learning and other computational batch workloads as customers can harness the power of GPUs to run distributed batch workloads at predictably affordable prices,” the company wrote in a blog post.
Google also provides a way for developers to automatically set up new preemptible virtual machines with GPUs through its managed instance groups feature. Preemptible virtual machines are combined with cloud products existing atop Compute Engine like Kubernetes Engine. The company said the GPUs are currently available in the US-central1 region.
