Benchmarking Amazon EC2's New C6a Instances Powered By 3rd Gen EPYC
For Amazon's own Arm processors there is C6g as their Graviton2-based instances. Amazon already announced Graviton3 with C7g instances coming but as of testing they aren't yet available and we haven't any access to them early yet from AWS. In any case Graviton2 is some healthy competition still for AMD EPYC 7003 series with having up to 64 Arm Neoverse N1 cores at cock speeds up to 2.5GHz and eight channels of DDR4-3200 memory.
The C6g.8xlarge instance offers 32 vCPUs and 64GB of RAM like the C6a.8xlarge. The C6g.8xlarge though does enjoy significantly lower cloud costs thanks to being Amazon's in-house processor. The C6g.8xlarge on-demand instance rate in Oregon is $1.088 to the C6a.8xlarge at $1.224 USD per hour.
It largely comes down to the workload though whether C6g instances make sense -- primarily how well optimized your in-use software is for AArch64 (or in cases of proprietary software, if AArch64 is even an option). The Graviton2 C6g instances can perform incredibly well if the workload is properly tuned for AArch64 while in other cases can deliver comparable (or sometimes worse) performance than the C6a. The C6g instances do carry the advantage of each vCPU being backed by a physical Neoverse N1 core compared to the AMD/Intel EC2 instances relying on their vCPUs as a mix of physical cores and the sibling SMT (HT) thread. Depending upon your needs and deployment approach, the C6g instances do only scale up to c6g.16xlarge at 64 vCPUs with Graviton2 only allowing 64 cores per server while the C6a Zen 3 instances can go beyond that with 96, 128, and 192 vCPU options.