Analysis: PostgreSQL Performance & Cost Across 23 AWS EC2 Instance Types

Source: The Sane Engineer Objective: Provide a data-driven framework for sizing PostgreSQL servers on AWS by balancing data volume, required throughput, and budget.

~~Stop guessing your instance sizes.~~ Instead, use benchmarked data to find the intersection of performance and price.

🛠️ Benchmark Methodology

The tests were designed to simulate a realistic production environment with the following parameters:

Database Engine: PostgreSQL 17
AWS Region: us-east-1
Pricing Model: On-demand
Concurrency: 32 simultaneous connections
Workload Distribution: $\text{Read} = 90\% \quad | \quad \text{Write} = 10\%$

The Selection Process

To choose the right instance, follow this logic flow:

Define your Dataset Size.
Define your Target Throughput (Requests Per Second).
Identify all instances that exceed that throughput.
Select the one with the lowest monthly cost.

📉 Case Study: 1 GB Dataset

For a small dataset of 1 GB with a target of 33,000 RPS, the results are as follows:

Recommendation Summary

Goal	Recommended Instance	Disk	Arch	Cost/Mo	Performance
Cheapest	`m8g.large`	`gp3-baseline`	Graviton	$82	45,515 RPS (1.4 $\times$ target)
Safe Pick	`m8g.large`	`gp3-baseline`	Graviton	$82	$\ge 20\%$ headroom
Max Efficiency	`c8i.large`	`gp3-baseline`	x86-64	~$84	52,203 RPS (Best RPS/$)

Requirement Checklist:

Clear 33,000 RPS
Minimize monthly spend
Maintain stability (headroom)

Note: 30 out of 52 tested configurations successfully cleared the 33,000 RPS threshold for a 1 GB dataset.

Figure 1: The dashed line represents the cost/performance frontier. Green points denote arm64 (Graviton) and slate points denote x86-64.

📊 Comprehensive Benchmark Results (1 GB)

The following table details the performance of various instance $\times$ disk combinations.

Instance Family	Arch	Disk	vCPU	RAM (GiB)	RPS	RPS/$	$/hr	$/mo	Avg ms	p95 ms	p99 ms	RPS/vCPU	RPS/GiB	Headroom
`c8i.large`	x86-64	`gp3-baseline`	2	4	52,203 $\pm 0.4\%$	451,476	$0.1156	~$84	0.61	2.16	2.79	26,102	13,051	1.6 $\times$
`m8g.large`	arm64	`gp3-baseline`	2	8	45,515 $\pm 0.1\%$	407,555	$0.1117	~$82	0.70	2.02	2.76	22,757	5,689	1.4 $\times$
`m8g.xlarge`	arm64	`gp3-baseline`	4	16	70,871 $\pm 0.1\%$	351,826	$0.2014	~$147	0.45	1.83	2.32	17,718	4,429	2.1 $\times$
`m7i.large`	x86-64	`gp3-baseline`	2	8	41,305 $\pm 0.4\%$	336,586	$0.1227	~$90	0.77	2.33	3.11	20,653	5,163	1.3 $\times$
`m7g.xlarge`	arm64	`gp3-baseline`	4	16	58,726 $\pm 0.4\%$	317,234	$0.1851	~$135	0.54	1.98	2.44	14,681	3,670	1.8 $\times$
`c8i.xlarge`	x86-64	`gp3-baseline`	4	8	64,585 $\pm 0.1\%$	308,522	$0.2093	~$153	0.50	1.85	2.27	16,146	8,073	2.0 $\times$
`r8g.large`	arm64	`gp3-baseline`	2	16	38,531 $\pm 1.6\%$	304,746	$0.1264	~$92	0.83	2.35	3.06	19,266	2,408	1.2 $\times$
`m7i.xlarge`	x86-64	`gp3-baseline`	4	16	65,582 $\pm 0.2\%$	293,410	$0.2235	~$163	0.49	1.86	2.30	16,396	4,099	2.0 $\times$
`r8g.xlarge`	arm64	`gp3-baseline`	4	32	64,627 $\pm 0.6\%$	279,821	$0.2310	~$169	0.49	1.89	2.32	16,157	2,020	2.0 $\times$
`m8g.medium`	arm64	`gp3-baseline`	1	4	16,670 $\pm 0.4\%$	249,555	$0.0668	~$49	1.92	3.44	4.28	16,670	4,167	0.5 $\times$
`m7g.large`	arm64	`gp3-baseline`	2	8	24,669 $\pm 0.4\%$	238,304	$0.1035	~$76	1.30	2.83	3.58	12,334	3,084	0.7 $\times$
`r8g.medium`	arm64	`gp3-baseline`	1	8	16,482 $\pm 0.2\%$	222,197	$0.0742	~$54	1.94	3.51	4.29	16,482	2,060	0.5 $\times$
`m7g.2xlarge`	arm64	`gp3-baseline`	8	32	77,197 $\pm 1.6\%$	221,629	$0.3483	~$254	0.41	1.81	2.22	9,650	2,412	2.3 $\times$
`m8g.2xlarge`	arm64	`gp3-baseline`	8	32	81,451 $\pm 0.7\%$	213,807	$0.3810	~$278	0.39	1.75	2.13	10,181	2,545	2.5 $\times$
`c8i.large`	x86-64	`gp3-max`	2	4	53,444 $\pm 0.4\%$	211,563	$0.2526	~$184	0.60	2.17	2.83	26,722	13,361	1.6 $\times$
`c8i.xlarge`	x86-64	`gp3-max`	4	8	69,067 $\pm 0.2\%$	199,429	$0.3463	~$253	0.46	1.85	2.30	17,267	8,633	2.1 $\times$
`m8g.xlarge`	arm64	`gp3-max`	4	16	67,428 $\pm 0.2\%$	199,242	$0.3384	~$247	0.47	1.86	2.30	16,857	4,214	2.0 $\times$
`r8g.xlarge`	arm64	`gp3-max`	4	32	69,292 $\pm 0.2\%$	188,322	$0.3679	~$269	0.46	1.87	2.36	17,323	2,165	2.1 $\times$
`m5.xlarge`	x86-64	`gp3-baseline`	4	16	39,959 $\pm 0.7\%$	186,794	$0.2139	~$156	0.80	2.22	2.68	9,990	2,497	1.2 $\times$
`m7g.xlarge`	arm64	`gp3-max`	4	16	57,738 $\pm 0.1\%$	179,252	$0.3221	~$235	0.55	2.00	2.48	14,434	3,609	1.7 $\times$
`m7g.medium`	arm64	`gp3-baseline`	1	4	10,986 $\pm 0.1\%$	175,171	$0.0627	~$46	2.91	4.76	5.83	10,986	2,747	0.3 $\times$
`m7i.2xlarge`	x86-64	`gp3-baseline`	8	32	73,190 $\pm 0.5\%$	172,163	$0.4251	~$310	0.44	1.80	2.18	9,149	2,287	2.2 $\times$
`m5.large`	x86-64	`gp3-baseline`	2	8