Comparison
m7g vs m7i for PostgreSQL
Graviton3 (m7g) vs 4th-gen Intel (m7i) general-purpose EC2 instances on PostgreSQL — same generation, same sizes, measured throughput and cost.
Verdict
For this PostgreSQL workload, m7g.xlarge (arm64) is the best value at 293,214 req/s per $/hr — a 4% cost-efficiency lead over m7i.xlarge. The highest raw throughput comes from m7g.2xlarge at 75,741 req/s. Across architectures, the best Arm option (m7g.xlarge) leads the best x86 option (m7i.xlarge) by 4% on cost-efficiency.
Side by side
| Instance | Arch | $/hr | RPS | p95 ms | RPS/$ | RPS/vCPU |
|---|---|---|---|---|---|---|
| m7g.xlarge | arm64 | $0.1851/hr | 54,279 | 2.03 | 293,214 | 13,570 |
| m7i.xlarge | x86-64 | $0.2235/hr | 63,215 | 1.88 | 282,820 | 15,804 |
| m7g.2xlarge | arm64 | $0.3483/hr | 75,741 | 1.82 | 217,449 | 9,468 |
| m7i.2xlarge | x86-64 | $0.4251/hr | 71,930 | 1.81 | 169,201 | 8,991 |
| m7g.large | arm64 | $0.1035/hr | 5,466 | 11.85 | 52,802 | 2,733 |
| m7i.large | x86-64 | $0.1227/hr | 5,474 | 11.80 | 44,606 | 2,737 |
Measured at a 10 GB dataset on gp3-baseline storage. Open the sizing tool to change the dataset size, disk, or metric, or browse the full benchmark data.
Frequently asked questions
Which is better for PostgreSQL, m7g.large or m7i.large or m7g.xlarge or m7i.xlarge or m7g.2xlarge or m7i.2xlarge?
On cost-efficiency, m7g.xlarge leads at 293,214 req/s per $/hr. For raw throughput, m7g.2xlarge is fastest at 75,741 req/s. Across architectures, the best Arm option (m7g.xlarge) leads the best x86 option (m7i.xlarge) by 4% on cost-efficiency. Choose by whether you optimize for throughput per dollar or peak capacity.
m7g.xlarge vs m7i.xlarge: what's the difference for PostgreSQL?
m7g.xlarge delivers 54,279 req/s at 293,214 req/s per $/hr; m7i.xlarge delivers 63,215 req/s at 282,820 req/s per $/hr. That is a 4% cost-efficiency edge for m7g.xlarge on this workload.
What workload and configuration were these numbers measured on?
A mixed OLTP pattern — 90% primary-key reads, 10% single-row inserts — at 32 concurrent connections on PostgreSQL 17 in us-east-1, with a 10 GB dataset on a gp3-baseline gp3 volume. The benchmark harness is open source.