PostgreSQL Performance Showdown: Apple M1 vs. Intel Xeon Silver 4110

We recently conducted a performance comparison of PostgreSQL on two different architectures: an Apple M1 chip and an Intel Xeon Silver 4110. The goal was to evaluate the transaction processing speed and latency under a simple workload. This blog post details the setup, results, and some initial observations.

Test Setup:

We used pgbench, a standard PostgreSQL benchmarking tool, for this comparison. The following commands were executed on both systems:

pgbench -i testdb
pgbench -c 10 -t 1000 testdb

This initializes a database named testdb and then runs a benchmark with 10 clients, each performing 1000 transactions. We used the simple query mode for this test. The key parameters were:

• Number of clients: 10
• Number of threads: 1 (default)
• Transactions per client: 1000
• Query mode: simple

Results:

Here’s a comparison of the results:

Observations:

The results show a significant performance difference between the two systems. The Apple M1 outperformed the Intel Xeon Silver 4110 by a substantial margin. Specifically:

• Latency: The Apple M1 exhibited significantly lower latency (0.867ms) compared to the Intel Xeon (12.847ms). This suggests that the M4 can process individual transactions much faster.
• Transactions per Second (TPS): The Apple M1 achieved a dramatically higher TPS (11532.72 including connection time) compared to the Intel Xeon (778.37). This indicates the M4’s ability to handle a much greater volume of transactions per unit of time.

Initial Thoughts:

The massive difference in performance is striking. While a comprehensive analysis would require more in-depth investigation, a few potential factors contribute to this disparity:

• Architecture: The Apple M1’s unified memory architecture and high clock speed likely play a significant role in its superior performance. It’s designed for efficiency and low latency.
• Clock Speed vs. IPC: While the Intel Xeon has a higher clock speed, the Apple M1 likely has a much higher Instructions Per Cycle (IPC), meaning it can execute more instructions per clock cycle.
• Cooling and Power: The M1 is designed for a mobile environment, which often prioritizes power efficiency and thermal management. This could lead to more consistent performance under load.