Google Axion CPU With GCE C4A vs. AWS Graviton4 Performance

Last week Google announced the general availability of their C4A instances powered by their in-house Axion processors. I delivered launch-day benchmarks looking at the Google Axion CPU performance with the C4A instances compared to other Google Cloud instance types powered by Ampere Altra and Intel Xeon. In this article is a look at how the Google Axion processor performance compares to the competing Amazon/AWS Graviton4 processor.

Both the Google Axion and AWS Graviton4 processors are based on Arm Neoverse-V2 designs. Graviton4 debuted earlier this year and features up to 96 cores, 2MB of L2 cache per core, and 12 channel DDR5-5600 memory. Graviton4 powers the Amazon Web Services EC2 R8g instances among other 8th gen instance types. Google Axion with the C4A machine series offers up to 72 vCPUs while the DDR5 memory details haven't been published by Google.

For getting a quick idea of the Graviton4 vs. Axion performance, I ran benchmarks of the AWS R8g instance type at 48 vCPUs and then of C4A Axion at 48 vCPUs within Google Cloud. With the C4A instance I tested both in the standard size (192GB of DDR5 memory for 48 vCPUs) and then in the "highmem" size for 384GB of memory to match that of the R8g.12xlarge instance.

Both the AWS Graviton4 and Google Axion instances were tested using Ubuntu 24.04 LTS at its defaults and with 215GB of storage. This comparison is just looking at the raw CPU performance between Graviton4 and Axion with no CPU power metrics exposed to the VMs or other means of being able to compare the real CPU or system power efficiency / performance-per-Watt.

This is a quick and straight-forward comparison so let's jump straight to the data. Thanks to Google for allowing the free/gratis access to Axion/C4A for benchmarking on Phoronix.