Google enhances its infrastructure-as-a-service (IaaS) platform with the launch of preemptible GPUs.
The preemptible GPUs, according to Google, is more affordable way for enterprises to run artificial intelligence workloads in public cloud. The users can attach Nvidia Tesla K80 and Nvidia Tesla P100 to Preemptible VMs (virtual machines), both of which are available on Google Cloud Platform.
The search engine giant charges only $0.22 and $0.73 per hour for Nvidia K80 and Nvidia P100, respectively. It will enable customers to harness the power of GPUs and run distributed batch workloads at minimum cost. Specifically, the Preemptible GPUs will be more appropriate for heavy machine learning workloads.
“Just like its on-demand equivalents, preemptible pricing is fixed. You’ll always get low cost, financial predictability and we bill on a per-second basis. Any GPUs attached to a Preemptible VM instance will be considered Preemptible and will be billed at the lower rate,” Google CE Product Managers wrote in a blog post.
You can use these GPUs continuously for 24 hours, and Google gives you only a 30-second warning to shut down your GPUs safely. As a result, the Preemptible GPUs are better for only those workloads that don’t need accessibility for long time.
Google has integrated the Preemptable VMs into cloud products which are built on compute engines. The support for Google Kubernetes Engine’s GPU is currently in preview.
The preemptible resources have been used widely by customers in all fields including satellite image analysis, financial services, quantum physics, to name a few. The Preemptible GPUs will increase the work productivity of customers, and reduce the costs by around 50%.
Additionally, the GPUs are now available in Google’s US-Central1 region. The support for Nvidia P100 has been released in Beta with no service-level agreement in any zone, while support for Nvidia K80 is now generally available.