MacBook Neo rival cloud servers in database load testing

In an interesting test, Gábor Szárnyas of DuckDB compared a 512GB MacBook Neo to a range of cloud servers to see how Apple’s new entry-level laptop performs under a heavy database load. Here’s how it happened.

MacBook Neo goes up against cloud servers with up to 4× more memory

In a blog post titled Big Data on the Cheapest MacBook (with Boing Boing), Szárnyas explains how he measured the MacBook Neo using two benchmarks: ClickBench and TPC-DS:

ClickBench has 43 questions that focus on grouping and sorting functions. The jobs run on one wide table with 100M rows, using about 14 GB when generated in Parquet and 75 GB when stored in CSV format.

The TPC-DS has 24 tables and 99 questions, many of which are complex and include features such as windowing. And while the TPC-H is designed to die for, there are still some price similarities in the results of the TPC-DS.

In all tests, the MacBook Neo faced two cloud conditions:

• c6a.4xlarge 16 AMD EPYC vCPU cores and 32 GB RAM.
• c8g.metal-48xl with 192 cores Graviton4 vCPU and 384 GB RAM.

In the ClickBench benchmark, they performed two tests: a cold run, which measures performance when the caches are empty, and a hot run, which measures performance when the system can take advantage of temporary storage.

In cold performance, the MacBook Neo beats both cloud conditions significantly, completing all queries in less than a minute, up to 2.8 times faster than its counterparts.

Although impressive, DuckDB explains that:

Of course, if we go deeper into the setup, there is an explanation for this. Cloud environments have network-attached disks, and database access to these controls the runtime of the entire query. The MacBook Neo has a local NVMe SSD, which is far from the best, but still provides fast access to the first read.

Things changed during the hot run test: the c8g.metal-48xl finished the run in 4.35 seconds, the c6a.4xlarge came in as the runner-up in 47.86 seconds, and the MacBook Neo finished last in 54.27 seconds, about 10% faster than in the cold.

However, it’s worth noting that in median query runtimes the MacBook Neo can still beat c6a.4xlarge, a medium-sized cloud instance. And the laptop’s overall runtime is 13% slower despite the cloud box having 10 CPU threads and 4 times more RAM.

As for the TCP-DS benchmark, DuckDB provides little comparative information, but it shows that the MacBook Neo still performs very well, considering its hardware:

On the SF100, the laptop breezed through most queries with an average query run time of 1.63 seconds and a total run time of 15.5 minutes.

In the SF300, the memory limitation started to appear. While the average query run time was still quite good at 6.90 seconds, DuckDB occasionally used up to 80 GB of disk space and it was obvious that some queries would take longer. Most notably, question 67 took 51 minutes to complete. But the hardware and software continued to work tirelessly together, and they finally passed the test, finishing all the questions in 79 minutes.

Interestingly, this was not the first time they tested the A19 Pro chip. Back when the iPhone 16 Pro came out, they ran the TCP-H benchmark with the device inside a bucket of dry ice at -50ºC, where it finished in 478.2 seconds.

To read more about DuckDB benchmarks on MacBook Neo, follow this link.

It’s worth checking out on Amazon