hpc7a.24xlarge

Analysis

The hpc7a.24xlarge is a compute optimized instance in the hpc7a series with 48 vCPUs and 768 GiB of memory on AMD EPYC 9R14 (up to 3.7 GHz), which makes it very memory-heavy at 16 GiB of memory per vCPU.

Running it in Ohio instead of Ireland saves $383 a month per instance: $7.20/hour against $7.73/hour, a 7.3% spread. It is not offered in N. Virginia, Oregon, Canada Central, São Paulo, London, Frankfurt, Mumbai, Singapore, Sydney and Tokyo, so plan around that if you need a presence there.

Per vCPU it is one of the dearer compute optimized types: 100% of the 291 current types in the family cost less than its $0.1500 per vCPU-hour. The premium mostly pays for the extra memory per vCPU. On demand it is $7.20/hour in its cheapest tracked region, or $5,256 for a 730-hour month.

It is size 2 of 4 in the hpc7a series. Stepping up to the hpc7a.48xlarge gives 96 vCPUs and 768 GiB for $0.00000 more a month. Dropping to the hpc7a.12xlarge saves $0.00000 a month. Unit price is not flat across the series ($0.0375 to $0.3000 per vCPU-hour), so the size changes the rate you pay.

Looking beyond its own line, 3 of 4 current types with 48 vCPUs and similar memory are cheaper. The cheapest of those is the x2gd.12xlarge on AWS Graviton2 at $4.01/hour.