In an intriguing experiment, DuckDB’s Gábor Szárnyas assessed the 512GB MacBook Neo alongside various cloud servers to evaluate how Apple’s latest budget laptop handles demanding database tasks. Here’s the outcome.
### MacBook Neo faces off against cloud servers with up to 4× greater memory
In a blog entry titled [Big Data on the Cheapest MacBook](https://duckdb.org/2026/03/11/big-data-on-the-cheapest-macbook) (via [Boing Boing](https://boingboing.net/2026/03/13/apples-cheapest-laptop-beat-a-cloud-server-with-4x-the-memory.html)), Szárnyas details how he evaluated the [MacBook Neo](https://amzn.to/47vJmkn) using two benchmarks: ClickBench and TPC-DS:
> ClickBench consists of 43 queries concentrating on aggregation and filtering tasks. These operations execute on a single extensive table containing 100M rows, which uses approximately 14 GB when serialized into Parquet and 75 GB when kept in CSV format.
> TPC-DS involves 24 tables and 99 queries, many of which are more intricate and feature elements like window functions. Although TPC-H has been excessively optimized, TPC-DS results still hold some degree of significance.
In every test, the [MacBook Neo](https://amzn.to/47vJmkn) was compared against two cloud instances:
– c6a.4xlarge featuring 16 AMD EPYC vCPU cores and 32 GB RAM.
– c8g.metal-48xl equipped with an exceptional 192 Graviton4 vCPU cores and 384 GB RAM.
For the ClickBench benchmark, they executed two tests: a cold run, assessing performance with empty caches, and a hot run, measuring performance once the system benefits from caching.
In the cold run, the [MacBook Neo](https://amzn.to/47vJmkn) significantly outperformed both cloud instances, completing all queries in under a minute, achieving speeds up to 2.8 times faster than its competitors.
While this is remarkable, DuckDB clarifies that:
> Naturally, if we examine the setups more closely, an explanation emerges. The cloud instances utilize network-attached disks, and accessing the database on these influences the overall query durations. The MacBook Neo employs a local NVMe SSD, which, while not top-tier, still offers fairly rapid access during the initial read.
The scenario shifted during the hot run test: c8g.metal-48xl completed the test in 4.35 seconds, c6a.4xlarge followed as a distant second at 47.86 seconds, and the [MacBook Neo](https://amzn.to/47vJmkn) came in last at 54.27 seconds, about 10% quicker than in the cold run.
Notably, regarding median query durations, the [MacBook Neo](https://amzn.to/47vJmkn) can still outperform the c6a.4xlarge, which is a mid-range cloud instance. Additionally, the total runtime of the laptop is only roughly 13% slower, despite the cloud instance possessing 10 additional CPU threads and four times the RAM.
Concerning the TCP-DS benchmark, DuckDB provides somewhat less comparative information but indicates that the [MacBook Neo](https://amzn.to/47vJmkn) still performed admirably given its specifications:
> At SF100, the laptop navigated most queries effortlessly with a median query runtime of 1.63 seconds and a total runtime of 15.5 minutes.
> At SF300, the limitations of memory began to surface. Although the median query runtime remained impressive at 6.90 seconds, DuckDB occasionally consumed up to 80 GB of space for disk spilling, and it became apparent that some queries would require extended time. Notably, query 67 took 51 minutes to finish. However, hardware and software continued to collaborate effectively, ultimately completing all queries in 79 minutes.
Interestingly, this was not the first instance of testing the A19 Pro chip. When the iPhone 16 Pro was released, they [conducted the TCP-H benchmark](https://duckdb.org/2024/12/06/duckdb-tpch-sf100-on-mobile#a-song-of-dry-ice-and-fire) with the device submerged in a bucket of dry ice at -50ºC, where it finished the run in 478.2 seconds.
To discover more about DuckDB’s benchmarks on the MacBook Neo, [follow this link](https://duckdb.org)