Google today announced that it has made graphics processing units (GPUs) available in beta through the Google Compute Engine. Specifically, Nvidia’s Tesla K80 GPUs can be attached to regular or custom virtual machine (VM) instances. Google first announced that it would make GPUs available through the Google Cloud Platform (GCP) in November.
As of today, you can only provision a GPU through Google’s gcloud software development kit (SDK). That option will become available through the GCP Console “later this week,” Google product manager John Barrus wrote in a blog post.
However, it turns out that if you’d like to try attaching a GPU to a VM through the gcloud command line interface (CLI) right now, you’ll encounter an error that looks something like this:
ERROR: (gcloud.beta.compute.instances.create) Some requests did not succeed:
Quota 'NVIDIA_K80_GPUS' exceeded. Limit: 0.0
In other words, by default, you cannot start using any GPUs. You have to manually request an increase in your quota using a Google Form on a per-region basis. Documentation does suggest that this is the case, and a Google spokesperson confirmed it to VentureBeat in an email. Today’s blog post does not acknowledge the quirk, though.
“Most established projects have quota already in place. Quota requests for new projects and free trials will be quickly filled (24-48 hours),” the spokesperson told VentureBeat in an email. “We are rolling out a console this week that will make the quota requests clear and easy.” Some existing projects already have GPU quotas, the spokesperson noted.
Public cloud market leader Amazon Web Services (AWS) offers GPU-backed instances but doesn’t let customers make custom VM instances. And it charges by the hour, while Google charges by the minute. The same can be said of Microsoft Azure’s N-Series GPU-backed instances.
While Google is bringing out GPUs after Amazon and Microsoft have done the same, Google is trying to stand out partly by offering a variety of GPUs to choose from. In the future, it will make AMD FirePro S9300 x2 GPUs available. The spokesperson wouldn’t comment on timing.