MMS • RSS
Article originally posted on InfoQ. Visit InfoQ
Google announced the general availability of GPUs in their Kubernetes Engine (GKE). Together with the recent GA of 1.10 version of GKE customers can land their machine learning (ML) workloads on to it and leverage the massive processing power of the GPUs.
Google offers several GPUs for GKE – a fast NVIDIA Tesla V100, a Tesla P100, and an entry-level Tesla K80. Each of these GPUs is available as Preemptible GPUs allowing customers to benefit from available GPUs in the Google Cloud at lower costs. Furthermore, with these GPUs available for GKE customers can benefit from some unique features according to the blog post on the announcement:
- Node Pools allow your existing cluster to use GPUs whenever you need
- Cluster Autoscaler automatically creates nodes with GPUs whenever pods are requesting GPUs are scheduled and scale down to zero when GPUs are no longer consumed by any active pods
- Taint and toleration technology ensures that only pods that request GPUs will be scheduled on the nodes with GPUs, and prevents pods that do not require GPUs from running on them
- Resource quota that allows administrators to limit resource consumption per namespace in a large cluster shared by multiple users or teams
Furthermore, Google also included the ability to monitor the GPU jobs to observe the performance through the GCP console.
Since the open-beta of GPUs on GKE, the number of core-hours increased by a ten-fold, indicating an increase in usage of Kubernetes on GCP. However, Google is not the only one seeing an increase in usage. Also, Microsoft saw an increase in usage of their Azure Kubernetes Service (AKS), which became generally available recently. Gabe Monroy, a project manager lead for containers at Microsoft Azure, said in a blog post early May of this year:
With Kubernetes exploding in popularity worldwide, it’s no surprise that Kubernetes usage on Azure has grown more than 10x over the last year.
Another significant public Cloud provider Amazon provides Elastic Container Service for Kubernetes (EKS), which is generally available since mid-June. Moreover, this provider leads in Kubernetes usage as a recent survey showed that 57% of companies running Kubernetes choose AWS.
Customers can start using the GPUs in Kubernetes Engine using the available free-trial of $300 credits. Furthermore, for price details of the GPUs see the pricing page.