Amazon EC2 M6i Performance For Intel Ice Lake In The Cloud Benchmarks

Overall, the new M6i instances from AWS are coming on strong. From the m5.24xlarge to m6i.24xlarge was a 36% improvement generationally with the same vCPU count, similar to the 33~38% improvement we were seeing with the 8xlarge instance type going from M5 to M6i. Considering that at least for now is the same on-demand pricing between M5 and M6i, this sixth-generation EC2 compute instance is quite an upgrade. However, we are still also waiting on Amazon to introduce their EPYC Milan instances (presumably "M6a") and will be very interesting to see how that compares to these Ice Lake instances in raw performance but also the value. AWS' latest indications are "later this year" for those EPYC 7003 instances.

Against Amazon's own Graviton2 instances, Ice Lake was competitive and often leading but largely comes down to the particular workloads of relevance to your needs for how well optimized they are for x86_64 (especially if AVX-512 usage is suitable) or tuned for AArch64. Graviton2 continues to show significant strength especially with all the instances being backed by physical cores for each vCPU and on a pricing basis does undercut the x86_64 instances.

Besides the m6i.24xlarge instance being around 36% faster than the m5.24xlarge instance for that former top-end configuration, if needing even greater performance out of a single EC2 instance there is now m6i.32xlarge for 128 vCPUs with 512GB of RAM. The m6i.32xlarge delivered a 64% improvement over the top-end Cascade Lake instance of m5.24xlarge or around 21% faster performance than the m6i.24xlarge, based on the geometric mean of all the benchmarks carried out.

Outside of performance, another benefit with the M6i instances are Ice Lake supporting hardware-based memory encryption (TME) to allow for a more secure public cloud.

Those interested in additional benchmark results and other metrics from this comparison can see all the data points over on OpenBenchmarking.org.