GPUs can also be used on Preemptible VMs by Google Cloud Platform (GCP) users, and now offers GPUs for Nvidia K80 and P100.
Although it is still a beta version, the release of this feature will bring another low-cost alternative to high computational effort such as high-throughput batch computing, machine learning, and scientific and technical computing. Compared to on-demand computing resources a lot cheaper, preemptive GPUs, the Nvidia K80 only 0.22 US dollars per hour, while the Nvidia P100 0.73 US dollars per hour.
Google introduced preemptive VMs in mid-2015, and preemptive instances are an affordable option as long as users' computing efforts are preemptive, as preempted VMs are fixed-priced, up to 80% cheaper than regular computing entities.
A preemptive VM can only exist for up to 24 hours, providing 30 seconds to save work progress when an instance is consumed for later operation, however, the Managed Instance Groups automatically rebuild the instance in the event of idle computing capacity. Computational resources are less stable, so preemptible VMs are only suitable for work that is fault-tolerant and does not require an exclusive instance.
Pre-emptive GPUs are used when additional GPUs are selected for preemptive VMs. Pre-emptive VMs for Google Cloud Platform are similar to Amazon EC2 Spot computing entities, but Spot computing entities can not attach GPUs. The reserved instances for AWS and Azure are also low The cost of computing options, but the need for longer-term contracts.