Google has announced the general availability of its C4A virtual machines, marking the debut of Axion-based instances. The cloud provider claims these instances deliver up to 10% better price-performance compared to the latest Arm-based alternatives from competitors, including Amazon Graviton4.
The C4A virtual machines are designed to handle general-purpose workloads, including web and app servers, microservices, databases, in-memory caches, media processing, and AI inference applications. These initial Axion-based instances are part of Google Cloud's general-purpose portfolio and are available in three configurations: Standard (1:4 vCPU-to-memory), High Memory (1:8 vCPU-to-memory), and High CPU (1:2 vCPU-to-memory). Salil Suri, director of product management at GCE, writes:
Google Cloud customers can use C4A in many services including Google Compute Engine, Google Kubernetes Engine (GKE), Batch, and Dataproc. C4A VMs are also now available in preview in Dataflow, with support for CloudSQL, AlloyDB, and other services coming soon.
The new VMs provide up to 50 Gbps of standard bandwidth and up to 100 Gbps with Tier 1 networking. They support the latest-generation Balanced and Extreme Hyperdisk storage, offering up to 350k IOPS and 5 GB/s throughput. In the recent article "Why Google keeps building custom silicon: The story behind Axion," Mark Lohmeyer, VP and GM at Google, and Parthasarathy Ranganathan, VP and technical fellow at Google, explain:
It was our experience of building such specialized chips not only for AI but also mobile and video streaming that gave us the confidence to tackle the more generalized though complex needs of CPUs (it sounds counterintuitive, but a general-purpose chip like Axion needs to handle a wider range of applications, which necessitates its more complex design).
Google is not the first cloud provider to introduce in-house designed Arm processors, with Axion following Amazon Graviton and Azure Cobalt. Reflecting the industry's shift toward Arm-based processors, Google claims its new Arm-based virtual machines offer up to 65% better price-performance and 60% greater energy efficiency than comparable current-generation x86-based instances.
Source: Google Cloud blog
The new instances leverage the Arm Neoverse V2 platform, and Ashok Bhat, senior product manager at Arm, emphasizes that Axion-based VMs outperform previous-generation Neoverse N1-based VMs and other x86 alternatives:
Google Cloud C4A VMs are an excellent choice for AI inference, capable of handling a wide range of workloads from traditional machine learning tasks like XGBoost to generative AI applications such as LLaMa.
Michael Larabel, founder of Phoronix.com, released the first independent performance benchmark of the Axion CPU, comparing it to existing GCE Arm and x86_64 instance types. Larabel also published a follow-up analysis comparing the new processor to Amazon’s competing Graviton4, providing the raw benchmark data for reference. Larabel writes:
Across the few dozen benchmarks conducted, while both Graviton4 and Axion are based on the Neoverse-V2 design, the C4A instance at 48 vCPUs was about 10% faster than the 48 vCPU Graviton4 instance using the r8g.12xlarge type.
Axion C4A instances are generally available through on-demand, Spot VMs, and various reservation and Committed Use Discount (CUD) options in a subset of regions, including Iowa, Virginia, Belgium, Frankfurt, and Singapore.