Google Cloud's Home Run in ARM-based VM space

Posted Oct 30, 2024

By Amulya Bhatia 4 min read

Audio summary

Google first announced Axion at their NEXT Event in April, which has gone in GA today. Axion is touted to outperform the ARM-based processors at the rival hyperscalers by at least 10% per vCPU on many workloads. It is based on Arm Neoverse™ V2 CPU, just like its counterpart AWS Graviton 4, whereas Azure Cobalt is relying on Arm Neoverse™ N2 Architecture.

In this article, I’ll shed some light on why the hyperscalers are making a massive push towards fabricating their own ARM-based Silicon and then top it off by some metrics for Axion and how you can migrate your workloads to it.

Historically, cloud servers have been predominantly powered by x86 architecture processors. However, with ARM-based processors becoming more powerful and efficient, hyperscalers like AWS, Azure, and Google Cloud are increasingly adopting ARM as a new standard. ARM is no longer just for mobile devices—it’s now powering the cloud.

Understanding ARM Architecture

x86 vs ARM

x86 is based on a Complex Instruction Set Computing (CISC) Architecture whereas ARM based processors follow Reduced Instruction Set Computing (RISC) Architecture.

Here’s a table comparing RISC and CISC architectures:

Feature	RISC (Reduced Instruction Set Computer)	CISC (Complex Instruction Set Computer)
Instruction Set	Small, highly optimized, simple instructions	Large, complex set of instructions
Instruction Complexity	Simple, each instruction performs a single task	Complex, each instruction can perform multiple operations
Execution Speed	Generally faster per cycle (often 1 cycle per instruction)	Slower per cycle; instructions may take multiple cycles to execute
Energy Efficiency	High efficiency, lower power consumption	Higher power usage due to complexity
Memory Usage	Can require more instructions, leading to potentially larger code	Fewer instructions per task, potentially saving memory
General Use Cases	Mobile devices, IoT, energy-efficient servers	Desktops, laptops, general-purpose servers
Advantages	Simpler design, lower power, faster per instruction	Fewer instructions per operation, potentially lower memory usage
Drawbacks	Higher memory usage due to more instructions	Higher power consumption, more complex design

In summary, ARM or Advanced RISC Machine, emphasizes performance-per-watt efficiency, which makes ARM processors highly power-efficient and cost-effective.

Why Hyperscalers Are Embracing ARM recently?

Cost Efficiency and Performance

ARM processors offer a balance of high performance at a lower cost than traditional x86 processors, which is attractive for hyperscalers. Their ability to handle massive parallel workloads efficiently can reduce operational costs.

Energy Efficiency and Sustainability

ARM’s power-efficient design can lead to significant reductions in energy consumption, which aligns with the sustainability goals of cloud providers. This energy efficiency is essential as the demand for cloud resources continues to grow.

See the chart below for comparative power consumption for a java application. For further context on this, read the full article here.

Customization and Control

ARM’s open licensing model allows cloud providers to customize processors for specific workloads.

Why did this shift take so long?

Primarily because of one reason: Software Compatibility. Most software was designed to run on x86, so a lot of work would be involved in hoarding everything over to ARM architecture. However, as it has increasingly gained popularity, it’s becoming much less of an issue. There will however still be applications which are optimized on assembler level which can’t be ported over to ARM, but most general-purpose applications have a much easier path in their ARM journey.

Comparison of ARM-Based VMs on Different Cloud Providers

The following table shows the current offerings - older AWS Graviton, Google Cloud Tau and Azure Ampere Altra were the precursors here.

Feature	AWS Graviton 4	Google Cloud Axion	Azure Cobalt
Processor Architecture	ARM Neoverse V2	ARM Neoverse V2	ARM Neoverse N2
Availability	Generally available	Generally available	Generally available
Key Use Cases	Application servers, midsize data stores, microservices,cluster computing, HPC, ML Inference, Gaming, Memory-cache	Web and App Servers, Databases, Gaming, Media Streaming, Ad Servers, BI & Data Analytics, ML Inference, Network Appliances, Memory-cache	Web applications, microservices, open-source databases, Memory-cache
Container/K8s Support	Yes	Yes	Yes
Main Instance Types	M8g, C8g, R8g	C4a (Standard, High-CPU, High-Mem editions available)	Dpsv6, Dplsv6, Epsv6
Region Availability	Select Regions	Select Regions	Select Regions

Axion

Axion is Google’s custom ARM-based processor with:

Upto 50% better performance and
Upto 60% better energy efficiency

compared to its x86 counterparts. For certain workloads, you can gain >100% efficiencies.

It allows upto 72 vCPUs, 576 gb RAM and 100 Gbps networking bandwidth. Configurations are available for the following CPU-to-Memory ratios - 1:2, 1:4 and 1:8 with Local SSD upto 6 TB.

Choosing the Right ARM-Based VM

When selecting an ARM-based VM, consider the following factors:

Workload Requirements: Assess the specific needs of your workload, such as CPU-intensive, memory-intensive, or I/O-bound tasks.
Operating System Compatibility: Ensure that your desired operating system is supported on the chosen ARM-based VM.
Pricing and Performance: Compare pricing and performance metrics across different cloud providers to find the best fit for your budget and requirements.
Support and Tools: Evaluate the availability of support resources, tools, and documentation for ARM-based VMs on each cloud platform.