Amazon EC2 M6i Performance For Intel Ice Lake In The Cloud Benchmarks
While there is a significant leap in performance to the new sixth-generation EC2 Intel (Ice Lake) instances over the fifth-generation with Cascade Lake, how does it compare to the sixth-generation AArch64 Graviton2 instances? That was the next area of focus for benchmarking... Primarily as we are still waiting to see AMD EPYC 7003 Milan instances in EC2, M6g with Graviton2 is the most technically interesting competition for the moment.
The m6g.8xlarge instance was used for continuing the look at the 32 vCPU size. It is important to note though that with the M6g "Graviton2" instances all vCPUs are backed by a physical AArch64 core rather than a mix of physical cores and sibling threads with the Intel (and AMD) EC2 instances.
With HPCG while there was a healthy 20% improvement from the Cascade Lake to Ice Lake instance, the m6g.8xlarge Graviton2 instance immediately jumped ahead... This is one of the areas where having each vCPU backed by a physical core really pays off.
Not only does the Graviton2 instance lead for raw performance with HPCG, but it also delivers much better value in this case. The m6g.8xlarge on-demand pricing is at $1.232 USD per hour compared to the Intel 8xlarge instances at $1.536.
For other HPC workloads, the m6g.8xlarge Graviton2 instance continued to outpace the m6i.8xlarge Ice Lake instance.
With AWS able to offer their own Graviton2 instances cheaper, it makes for great value in these workloads that play well in an AArch64 world.