Twice the price-performance of comparable modern x86 VMs is Google’s claim for its new N4A instances, now available in preview.
Google Cloud is doubling down on Arm-based computing with the introduction of Axion N4A, as it seeks to reduce dependency on x86-based instances and attract enterprises looking to optimize costs.
Its N series of compute instances are typically targeted at general purpose workloads such as low-to-medium traffic web and application servers; containerized microservices; virtual desktops; back-office, CRM, or BI applications; data pipelines; and small-to-medium databases.
The N4A instances, currently available in preview, are configurable up to 64 vCPUs with support for 512GB of DDR5 Memory and 50 Gbps networking connection.
Google is marketing N4A as the most cost-effective N series instance yet, offering “up to 2x better price-performance and 80% better performance-per-wattthan comparable current-generation x86-based VMs.”
The company is able to offer such cost savings because it designed the Arm-based Axion processors in-house, and sources them directly from manufacturers such as Taiwan Semiconductor Manufacturing Company (TSMC), effectively halving the cost of the underlying hardware, said Brandon Hoff, research director at IDC. Combined with the chips’ performance, this positions Google’s Arm-based instances as a compelling alternative for enterprises, he said.
The savings are so attractive that Google is already using Arm instances internally for services including YouTube, Gmail, and BigQuery.
It’s not alone: AWS introduced its own Arm-based chip, Graviton, in 2018 to reduce the cost of running internal cloud workloads such as Amazon retail IT, and now 50% of new AWS instances run on it. Microsoft, too, recently developed an Arm chip, Cobalt, to run Microsoft 365 and to offer Azure services.
Google’s N4A instances will be available across services including Compute Engine for running virtual machines directly, Google Kubernetes Engine (GKE) for running containerized workloads, and Dataproc for big data and analytics. They will be accessible in the us-central1 (Iowa), us-east4 (N. Virginia), europe-west3 (Frankfurt) and europe-west4 (Netherlands) regions initially.
The company describes N4A as a complement to the C4A instances it launched last October. These are designed for heavier workloads such as high-traffic web and application servers, ad servers, game servers, data analytics, databases of any size, and CPU-based AI and machine learning.
Also coming “soon” is C4A Metal, a bare-metal instance for specialized workloads in a non-virtualized environment, such as custom hypervisors, security workloads, or CI/CD pipelines.
