Amazon EC2 M6i Performance For Intel Ice Lake In The Cloud Benchmarks
In addition to the benchmarks looking at the performance across the same instance types, the second part of our initial M6i benchmarking is looking at the peak performance that can be achieved out of a single VM. For this testing the following instances were benchmarked for some follow-up tests:
m6g.metal - The largest instance currently offered with Graviton2 for having 64 vCPUs backed all by physical cores.
m5.24xlarge - The largest M5 Cascade Lake instance offered of 96 vCPUs (48 cores / 96 threads).
m6i.24xlarge - While not the largest Ice Lake instance, the m6i.24xlarge was also tested for matching the largest Cascade Lake instance. The m6i.24xlarge is using 48 cores / 96 vCPUs as well.
m6i.32xlarge - This is now the largest Ice Lake instance available in the M6i series at 128 vCPUs (64 physical cores). This instance also provides twice the RAM capacity of the m6g.metal instance.
The M6i Ice Lake instances are able to offer a lot more computing capacity than M6g or M5 for workloads that do not scale out as well to multiple instances. Both the 24xlarge and 32xlarge instances were much faster than the M6g.metal that taps out at 64 cores. The largest Ice Lake instance offered 1.82x the performance of the Graviton2 metal instance. For the Intel generational performance, going from m5.24xlarge to m6i.24xlarge was a 38% improvement in performance -- similar to our M5 to M6i 8xlarge benchmarking commonly in the 30%+ range.
When it comes to the on-demand pricing, the m6i.24xlarge was delivering the best value of the instances tested while the m6i.32xlarge was delivering similar value to the m6g.metal.
Going from m5.24xlarge to m6i.24xlarge was continuing to show around a 38% improvement with the same vCPU count and going from Cascade Lake to Ice Lake.