Google Cloud N4 Series Benchmarks: Google Axion Vs. Intel Xeon Vs. AMD EPYC Performance

Why It Matters

Google’s custom silicon proves it can compete with Intel and AMD on price‑performance, reshaping cloud‑provider hardware strategies and customer cost models.

Key Takeaways

• Axion N4A costs $0.71 per hour, cheapest.
• N4A matches or exceeds Xeon and EPYC performance.
• No SMT; 16 physical cores deliver high single‑thread throughput.
• Intel Xeon uses Emerald Rapids, not latest Granite Rapids.
• EPYC Turin instance slightly pricier but comparable performance.

Pulse Analysis

Google’s decision to design and fabricate its own Arm‑based Axion processor marks a strategic pivot from reliance on third‑party silicon. By integrating the chip into the N4A VM family, Google can tailor core architecture, cache hierarchies, and power envelopes to its hyperscale workloads. This vertical integration not only reduces dependency on external roadmaps but also opens the door for tighter software‑hardware co‑optimization, a competitive edge that cloud giants have long pursued through custom accelerators and specialized networking stacks.

The recent benchmark suite, conducted on identical 16‑vCPU, 400‑GB configurations running Ubuntu 25.10, reveals that the Axion‑powered N4A holds its own against Intel’s Emerald Rapids Xeon Platinum 8581C and AMD’s Zen 5 EPYC 9B45. Despite lacking simultaneous multithreading, the 16 physical cores deliver strong single‑thread performance, narrowing the gap that traditionally favored x86 CPUs in latency‑sensitive workloads. Multi‑core tests show comparable throughput, while the price‑per‑hour advantage—$0.71 versus $0.77 for EPYC and $0.82 for Xeon—translates into a clear performance‑per‑dollar win for many cloud customers.

For enterprises evaluating cloud spend, the N4A’s cost efficiency could shift workload placement decisions, especially for compute‑intensive, scale‑out applications that benefit from Arm’s power efficiency. Moreover, Google’s ability to iterate on processor design faster than Intel or AMD may accelerate feature rollouts such as enhanced security extensions or AI‑centric instructions. As other providers explore their own silicon strategies, the N4A benchmark underscores a broader industry trend: custom silicon is becoming a decisive factor in cloud pricing, performance, and ecosystem lock‑in.