Carbon-aware scheduling addresses when and where you run workloads. But there is a parallel lever: making the workload itself consume less electricity to do the same amount of work. This is hardware efficiency, and it has seen a dramatic shift in recent years.
The x86 Era
For decades, Intel and AMD x86 processors dominated data centers. They are powerful and highly compatible with existing software, but their architecture carries decades of legacy design decisions that prioritize raw performance over power efficiency. A typical x86 server CPU under load can draw 150–300 watts.
The ARM Revolution
ARM (Advanced RISC Machine) is a processor architecture originally designed for mobile devices, where battery life is paramount. ARM chips achieve high performance through a fundamentally different design philosophy: simpler instruction sets, more efficient pipelines, and aggressive power gating (turning off unused parts of the chip).
AWS designed their own ARM-based server chip called Graviton (currently in its third generation: Graviton3). Google Cloud offers Tau T2A instances based on Ampere Altra ARM processors. Azure offers Cobalt 100 ARM-based VMs.
For many common cloud workloads — web servers, containerized microservices, caching layers, API gateways — ARM-based instances deliver:
| Metric | x86 (m5.large) | ARM Graviton (m6g.large) |
|---|---|---|
| vCPU | 2 | 2 |
| RAM | 8 GB | 8 GB |
| On-demand price (us-east-1) | ~$0.096/hr | ~$0.077/hr |
| Relative performance | Baseline | +10–40% for many workloads |
| Energy per unit of work | Baseline | ~60% less |
The energy efficiency gain is not just about lower clock speeds — it is about the fundamental architecture doing more work per watt of electricity consumed.
The Migration Path
Migrating to ARM requires recompiling your application. For most modern languages this is straightforward:
- Go, Rust, C/C++: Recompile with
GOARCH=arm64or the equivalent compiler target. - Java, Python, Node.js: The runtime itself needs an ARM build (widely available), but your application code is often unchanged.
- Docker containers: Build a multi-architecture image using
docker buildxwith--platform linux/arm64. - Dependencies: The main risk is native binary dependencies (e.g., compiled C extensions in Python packages). Most popular packages now ship ARM64 wheels, but verify before migrating production workloads.
A practical GreenOps action with immediate impact: audit your largest compute workloads, identify those running on x86, and evaluate ARM migration. The combination of lower cost and lower energy use makes this one of the highest-ROI GreenOps changes available today.