Compute Engine, Google’s infrastructure-as-a-service (IaaS) offering that lets customers run workloads on a scalable number of cloud-hosted virtual machines (VMs), is becoming a bit less expensive. The Mountain View company today announced that it’s committing use discounts to support graphics cards (GPU), local solid state storage (SSD), and tensor processing unit (TPU) Pods, which are tailor-made for AI inferencing and training tasks like image recognition, natural language processing, and reinforcement learning.
If you aren’t familiar with use discounts, here’s a quick primer: Google extends bill credit to users for running Compute Engine resources a “significant portion” of the billing month. (For example, if someone were to run these resources for more than 25% of a month, Compute Engine would automatically give them a discount for every incremental minute they used.) Previously, use discounts applied only to processors, memory, and select GPU devices, but the list of eligible hardware is expanding substantially.
Google senior product manager Manish Dalwadi says the move is ultimately geared toward making Google Cloud pricing simpler, fairer, and more transparent. “You shouldn’t need an advanced degree in finance to get the most out of your cloud investment,” he wrote in a blog post, “and you definitely shouldn’t have to worry about your cloud provider covering up costs under layers of complexity.”
Starting this week in all Compute Engine regions (East Asia, Northeast Asia, South and Southeast Asia, Southeast Australia, Northern Europe, Southern Europe, Western Europe, U.S. East, U.S. West, Canada, and South America), users can purchase GPUs, TPU Pods, or local SSD storage for up to 55% off on-demand prices. They’ll have “total control” over the instance types, families, and zones to which they apply committed use discounts, Google says, in addition to a choice of GPUs — including Nvidia Tesla K80, P4, P100, and V100 and several slice sizes of Cloud TPU v2 Pods and Cloud TPU v3 Pods.
Google is also introducing capacity reservation for Compute Engine, which allows customers to reserve resources in a specific zone for later use. These reservations can be deleted at any time and consume resources just like normal VMs, so existing discounts (like sustained use discounts and committed use discounts) apply.
“Reservations help ensure you have compute capacity available when and where you need it and are especially useful for anticipated spikes, say, during the holiday, when performing backup and disaster recovery, or for planned organic growth,” wrote Dalwadi. “We will continue to introduce discounts and pricing options that are flexible and predictable … [and] we won’t make you choose from thousands of SKUs, or worse, hide costs in confusing and hard-to-use bundles.”