m6a.8xlarge

Analysis

The m6a.8xlarge is a general purpose instance in the m6a series with 32 vCPUs and 128 GiB of memory on AMD EPYC 7R13 (up to 3.6 GHz), which makes it balanced at 4 GiB of memory per vCPU.

It costs the same as the older m5a.8xlarge on AMD EPYC 7571, so the move up a generation is free. The newer m7a.8xlarge on AMD EPYC 9R14 is 34% dearer at $1.85/hour.

Running it in Mumbai instead of São Paulo saves $960 a month per instance: $0.8888/hour against $2.20/hour, a 148% spread.

Within the same generation and size, the m6g.8xlarge on AWS Graviton2 costs $1.23/hour, 11% less ($110 a month), though switching means rebuilding software for arm64. The other variants are the m6gd.8xlarge at $1.45, the m6i.8xlarge at $1.54 and the m6id.8xlarge at $1.90.

At $1.38/hour on demand in US East (N. Virginia), about $1,009 a month, the m6a.8xlarge is one of the cheaper ways to buy general purpose capacity: only 17% of the 354 current general purpose types cost less per vCPU-hour than its $0.0432.

Reserving it for a year with no upfront payment cuts the Linux rate to $0.9145/hour, which beats on demand once it runs more than 66% of the time; three years brings it to $0.6271/hour, 55% below on demand. Windows Server roughly doubles the bill at $2.85/hour, because the licence is charged per vCPU.