Google's Axion ARM CPU Benchmarked Against x86 Rivals
What happened
Phoronix benchmarked Google's ARM-based Axion CPU against Intel Xeon and AMD EPYC processors on the Google Cloud N4 platform. Results showed AMD was approximately 35% faster in some CPU-bound tasks, while Intel demonstrated a lead in memory and I/O performance, offering a mixed-performance landscape for datacenter chips.
Why it matters
- Google's Axion processor is built on Arm's Neoverse V2 CPU architecture, the same foundation used by competitors like Amazon for its Graviton4 CPU and Nvidia for its Grace CPU. - The Axion CPU is part of a broader system-on-a-chip architecture Google calls "Titanium," which uses dedicated microcontrollers to offload tasks like networking, security, and storage I/O processing, freeing up the main CPU cores for general-purpose workloads. - Google claims that Axion offers up to 50% better performance and up to 60% improved energy efficiency when compared to similar current-generation x86-based instances in its cloud. - The N4 platform referenced is a family of general-purpose instances; besides the Axion-powered N4A, Google Cloud also offers the N4 with Intel Xeon "Emerald Rapids" CPUs and the N4D with AMD EPYC "Turin" processors. - A broader Phoronix benchmark suite across 70 different tests found that, for a 16 vCPU instance, the AMD EPYC Turin-powered N4D was about 18% faster than the Axion-powered N4A overall. - In the same 16 vCPU comparison, the Axion N4A instance was roughly 18% faster than the Intel Xeon Emerald Rapids-powered N4 instance and had a lower hourly cost. - The performance difference in the 16 vCPU benchmarks is influenced by core topology; the Axion N4A instance utilized 16 physical cores, whereas the comparable AMD and Intel N4 instances used 8 physical cores with simultaneous multithreading (SMT) or Hyper-Threading. - This move makes Google the latest major cloud provider to design its own Arm-based data center CPUs, following Amazon's Graviton series and Microsoft's Azure Cobalt, signaling a major trend away from reliance on third-party x86 vendors.
Key numbers
- Phoronix benchmarked Google's ARM-based Axion CPU against Intel Xeon and AMD EPYC processors on the Google Cloud N4 platform.
- Results showed AMD was approximately 35% faster in some CPU-bound tasks, while Intel demonstrated a lead in memory and I/O performance, offering a mixed-performance landscape for datacenter chips.
- - Google's Axion processor is built on Arm's Neoverse V2 CPU architecture, the same foundation used by competitors like Amazon for its Graviton4 CPU and Nvidia for its Grace CPU.
- Google claims that Axion offers up to 50% better performance and up to 60% improved energy efficiency when compared to similar current-generation x86-based instances in its cloud.
Quick answers
What happened in Google's Axion ARM CPU Benchmarked Against x86 Rivals?
Phoronix benchmarked Google's ARM-based Axion CPU against Intel Xeon and AMD EPYC processors on the Google Cloud N4 platform. Results showed AMD was approximately 35% faster in some CPU-bound tasks, while Intel demonstrated a lead in memory and I/O performance, offering a mixed-performance landscape for datacenter chips.
Why does Google's Axion ARM CPU Benchmarked Against x86 Rivals matter?
Google's Axion processor is built on Arm's Neoverse V2 CPU architecture, the same foundation used by competitors like Amazon for its Graviton4 CPU and Nvidia for its Grace CPU. The Axion CPU is part of a broader system-on-a-chip architecture Google calls "Titanium," which uses dedicated microcontrollers to offload tasks like networking, security, and storage I/O processing, freeing up the main CPU cores for general-purpose workloads. Google claims that Axion offers up to 50% better performance and up to 60% improved energy efficiency when compared to similar current-generation x86-based instances in its cloud. The N4 platform referenced is a family of general-purpose instances; besides the Axion-powered N4A, Google Cloud also offers the N4 with Intel Xeon "Emerald Rapids" CPUs and the N4D with AMD EPYC "Turin" processors. A broader Phoronix benchmark suite across 70 different tests found that, for a 16 vCPU instance, the AMD EPYC Turin-powered N4D was about 18% faster than the Axion-powered N4A overall. In the same 16 vCPU comparison, the Axion N4A instance was roughly 18% faster than the Intel Xeon Emerald Rapids-powered N4 instance and had a lower hourly cost. The performance difference in the 16 vCPU benchmarks is influenced by core topology; the Axion N4A instance utilized 16 physical cores, whereas the comparable AMD and Intel N4 instances used 8 physical cores with simultaneous multithreading (SMT) or Hyper-Threading. This move makes Google the latest major cloud provider to design its own Arm-based data center CPUs, following Amazon's Graviton series and Microsoft's Azure Cobalt, signaling a major trend away from reliance on third-party x86 vendors.
