Google has become the first major cloud provider to start offering virtual machine instances using a new generation of Intel Xeon processors powered by the chipmaker’s Skylake architecture. The company launched those instances in beta a few months ago, and they’re generally available as of Wednesday.
It’s a move by Google to better compete with other cloud providers like Amazon Web Services and Microsoft Azure, which have yet to launch their own Skylake infrastructure products. AWS announced a set of Skylake-powered C5 instances for its Elastic Compute Cloud last year, but those have yet to launch.
In addition, Google has made several other updates to its Compute Engine infrastructure-as-a-service offering. Virtual machine instances running in GCE can now support up to 64 vCPUs and 455GB of RAM. Using a drop-down menu, developers can choose their minimum preferred processor architecture for a particular Google Cloud region. The company offers a variety of Intel Xeon E5 processors, ranging from Intel’s Ivy Bridge architecture up to Skylake.
As of Wednesday, developers can finally run workloads on Broadwell chips in every Google Cloud region, though not every data center within those regions support them.
Skylake instances are available in three of Google’s eight cloud regions: Western US (Oregon), Western Europe (Belgium), and Eastern Asia Pacific (Taiwan). The company says that it plans additional geographic availability soon.
For the first 60 days of general availability, Skylake instances will be priced the same as their older counterparts. After that, users will be charged a premium price for using them — roughly 6-10 percent above the non-Skylake price, depending on the instance configuration.
Google has also lifted restrictions on how much memory can be connected to a single virtual CPU for customers who choose to customize their virtual machine instances. Customers can now elect to use as much as 455GB of memory with their VMs, no matter how many vCPUs those machines have.
Previously, developers were restricted to only using 6.5GB of RAM per vCPU, which meant workloads that need a lot of memory could end up requiring more processing power than they needed. Using more than 6.5GB of RAM per vCPU will incur an extra surcharge, however.