Google Cloud’s newest Arm-based virtual machine instances are delivering performance that should give both AMD and Intel pause. In extensive benchmarking conducted by Phoronix, the Google Axion-powered N4 Arm64 instances demonstrated compelling price-performance advantages across a wide range of workloads, signaling that the Arm architecture’s march into the data center is no longer a future prospect — it is the present reality.
The tests, carried out by veteran Linux benchmarking expert Michael Larabel, compared Google Cloud’s N4 Arm64 instances against the C4 instances powered by Intel’s latest Xeon Scalable processors (Emerald Rapids) and the C3D instances running AMD’s EPYC (Genoa) chips. All instances were configured with comparable vCPU and memory allocations, providing a controlled environment to measure raw computational throughput and, perhaps more critically, the cost efficiency that cloud customers actually care about.
Google Axion: A Custom Arm Chip Built for Cloud Workloads
The N4 instances are powered by Google Axion, a custom Arm-based processor that Google designed in-house using Arm’s Neoverse V2 cores. Announced in April 2024 and made generally available later that year, Axion represents Google’s direct answer to AWS Graviton and Ampere Altra chips that have been gaining traction across cloud providers. Google has claimed that Axion delivers up to 50% better performance and up to 60% better energy efficiency compared to comparable x86-based instances across certain workloads.
The Phoronix review tested these claims with a battery of over 100 benchmarks spanning compilation tasks, database operations, encryption, compression, scientific computing, and web serving. The results paint a nuanced picture: while Axion does not dominate every single test, its aggregate performance — especially when factored against Google Cloud’s per-hour pricing — tells a story that enterprise architects and cloud cost engineers cannot afford to ignore.
Raw Performance: Where Arm Wins and Where x86 Still Holds Ground
In the Phoronix testing, the N4 Arm64 instances showed particular strength in multi-threaded workloads. Compilation benchmarks — a proxy for general-purpose compute throughput — favored the Arm instances in several configurations. The Linux kernel compilation test, a widely recognized benchmark in the open-source community, showed the N4 Arm64 instances completing builds faster than both the Intel-based C4 and AMD-based C3D instances at comparable vCPU counts.
Database and web-serving benchmarks told a similar story. In PostgreSQL pgbench tests, the Arm instances posted strong transaction-per-second numbers. Redis and Memcached benchmarks, which stress memory subsystem performance and low-latency data access, also showed the N4 instances performing competitively. The Arm architecture’s memory bandwidth characteristics, combined with Google’s custom silicon tuning, appear to give Axion an edge in workloads that are sensitive to memory throughput.
AMD EPYC Genoa Fights Back on Single-Threaded Tasks
AMD’s EPYC Genoa processors, powering the C3D instances, did not go quietly. In several single-threaded and lightly-threaded benchmarks, the Zen 4 architecture’s high IPC (instructions per clock) and clock speed advantages allowed the C3D instances to match or beat the Arm-based N4 machines. Workloads involving heavy floating-point computation and certain encryption algorithms also favored the x86 chips, reflecting the maturity of x86 SIMD instruction sets like AVX-512.
Intel’s Xeon-based C4 instances, meanwhile, showed respectable performance but frequently trailed both AMD and Arm in the Phoronix tests. The Emerald Rapids generation, while an improvement over its predecessors, has struggled to match AMD’s Genoa in many data center benchmarks — a trend that has been documented across multiple independent reviews over the past year. The addition of a competitive Arm option from Google only compounds Intel’s challenge in the cloud compute market.
The Price-Performance Equation Changes the Calculus
Perhaps the most significant finding from the Phoronix analysis is not about raw speed at all — it is about cost. Google prices its N4 Arm64 instances at a discount compared to the x86 alternatives. According to the Phoronix review, when performance is normalized against the per-hour cost of each instance type, the N4 Arm64 instances frequently deliver the best value proposition. In several benchmark categories, the price-performance advantage of the Arm instances exceeded 30% compared to the Intel C4 instances and roughly 20% compared to the AMD C3D instances.
This pricing dynamic mirrors what Amazon Web Services has achieved with its Graviton processor family. AWS has consistently priced Graviton instances below comparable x86 offerings, and customers who have recompiled their workloads for Arm have reported meaningful cost reductions. Google appears to be following the same playbook, using its own custom silicon to offer lower prices while maintaining or improving margins — a strategy that puts direct pricing pressure on AMD and Intel, whose chips carry licensing and manufacturing costs that custom Arm designs can sometimes avoid.
Software Compatibility: The Remaining Friction Point
The transition to Arm in the cloud is not without friction. While the Linux kernel and most major open-source software stacks have excellent Arm64 support in 2025, enterprise applications and certain proprietary software packages may still require x86. The Phoronix benchmarks were conducted entirely on Linux with open-source toolchains, which represents the best-case scenario for Arm compatibility.
That said, the software gap has narrowed dramatically. Major databases (PostgreSQL, MySQL, MongoDB), container runtimes (Docker, containerd), orchestration platforms (Kubernetes), and programming language runtimes (Java, Python, Go, Node.js, Rust) all have mature Arm64 support. Google has invested in ensuring that its managed services — including Google Kubernetes Engine (GKE) — work natively with the N4 Arm64 instances. For organizations running containerized microservices, the migration path from x86 to Arm often requires little more than rebuilding container images for the new architecture.
A Three-Way Race With Broader Industry Implications
The benchmarking results underscore a broader transformation underway in the cloud computing hardware market. For more than two decades, Intel dominated the server processor business with minimal competition. AMD’s resurgence with the Zen architecture beginning in 2017 disrupted that dominance, and now Arm-based processors from cloud providers themselves are adding a third vector of competition. Each of the three major hyperscalers — Amazon, Google, and Microsoft — now either has or is developing custom Arm-based server chips.
Microsoft’s Cobalt 100 processor, based on the same Arm Neoverse V2 cores as Google Axion, has begun appearing in Azure virtual machines. Amazon’s Graviton4, the latest in its custom chip line, is already available across multiple AWS instance families. This convergence around Arm Neoverse designs means that software optimized for one cloud provider’s Arm instances will generally run well on another’s, reducing lock-in concerns and accelerating adoption.
What This Means for Enterprise Cloud Spending
For enterprise IT leaders evaluating cloud infrastructure costs — which Gartner has estimated will exceed $700 billion globally in 2025 — the arrival of competitive Arm-based options introduces a meaningful new variable into procurement decisions. A 20-30% improvement in price-performance, as suggested by the Phoronix data, can translate into millions of dollars in annual savings for large-scale deployments.
The decision is not purely financial, of course. Operational considerations — including team expertise, existing CI/CD pipelines, and vendor support agreements — all factor into architecture choices. But the performance data is increasingly hard to argue with. As Phoronix noted, the Google Cloud N4 Arm64 instances represent a strong option for a wide range of general-purpose and scale-out workloads, and their pricing makes them particularly attractive for cost-conscious deployments.
The Pressure Mounts on Traditional Chip Makers
For AMD and Intel, the message from these benchmarks is clear: the competitive threat from custom Arm silicon in the cloud is real and growing. AMD’s EPYC line remains highly competitive on absolute performance, but it faces pricing pressure from chips that cloud providers design and manufacture on their own terms. Intel, already losing market share to AMD in the data center, now faces a two-front challenge that its upcoming Granite Rapids and Sierra Forest architectures will need to address decisively.
The cloud computing market has entered a period of genuine three-architecture competition — x86 from Intel, x86 from AMD, and Arm from the hyperscalers themselves. For customers, this competition is unambiguously positive, driving down costs and pushing performance forward. For the chip makers, the stakes have never been higher. The benchmarks from Google Cloud’s N4 instances are just the latest data point in a competitive dynamic that will define the server processor market for years to come.