Google Cloud Now Offers Low-Cost Preemptible GPUs

Businesses using Google’s cloud computing services for resource hungry GPU processes can now avail of a lower cost option recently launched by the company. To sweeten the deal, the new feature, which was already launched as a beta version, is offered to companies at a significant discount. However, there is a catch—Google has the right to shut down any process anytime if it needs those resources.

Called preemptible GPUs, clients can now create GPU instances in the Google Cloud Platform using Nvidia chips. The company announced that these GPU processes are priced 50 percent lower than GPUs attached to on-demand instances, which could be big savings for clients and especially attractive for firms pressured to reel in operational expenses.

But of course, this 50 percent discount comes with a tradeoff. Google may, at any time, terminate these preemptible GPU instances if it needs these resources. As explained in by the tech giant, the compute engine will terminate them after giving a 30-second warning. In addition, Preemptible GPUs may only be used for a maximum of 24 hours.

Under the preemptible feature, clients will have access to Nvidia K80 and Nvidia P100 GPU at a much lower rate. The company will charge K80 usage at $0.22 per GPU hour while the P100 carries a $0.73 per hour rate.

With the preemptible option, Google is targeting clients running “distributed and fault-tolerant workloads,” especially processes that are dependent on a single instance.  Another advantage seen from this option is clients will know exactly how much they are going to pay compared to the currently unpredictable pricing models for on-demand services.

At the moment, Google announced that the preemptible GPU feature is now available in the US-central1 region. To check if the feature is already available in your region, click the link for the full list of locations.

[Featured image via Google Cloud]