MacBook Neo Challenges Cloud Servers in Database Test

A recent performance evaluation pitted Apple’s 512GB MacBook Neo against powerful cloud server instances in a series of demanding database benchmarks. The results reveal that Apple’s entry-level laptop can deliver surprisingly competitive performance in specific scenarios, challenging the assumption that serious computational work requires expensive remote infrastructure. The tests, conducted using the DuckDB analytical database, provide a fascinating look at how modern, integrated hardware can sometimes outperform more traditionally powerful setups.

The experiment compared the MacBook Neo against two Amazon Web Services cloud instances. The first was a c6a.4xlarge instance with 16 AMD EPYC vCPU cores and 32 GB of RAM. The second was a far more formidable c8g.metal-48xl instance equipped with 192 Graviton4 vCPU cores and a massive 384 GB of RAM. The laptop faced these servers using two standard benchmarks: ClickBench and TPC-DS.

ClickBench focuses on aggregation and filtering operations across a single large table containing 100 million rows. The team ran both a “cold” test with empty caches and a “hot” test that allowed systems to benefit from cached data. In the cold run, the MacBook Neo’s performance was startling. It finished all queries in under a minute, beating both cloud servers by a significant margin,up to 2.8 times faster.

This initial advantage has a clear explanation. The cloud instances rely on network-attached storage, where accessing the database files dominates the total query time. In contrast, the MacBook Neo uses a local NVMe solid-state drive. While not the fastest available, this local storage provides much quicker initial data access, giving the laptop a substantial head start when data is not already cached in memory.

The situation reversed in the hot run test, where data could be read from memory. Here, the powerful c8g.metal instance completed the workload in just 4.35 seconds. The c6a.4xlarge instance finished in 47.86 seconds, and the MacBook Neo came in last at 54.27 seconds. Despite this, the laptop’s median query runtime was still faster than the mid-sized cloud server. Its total runtime was only about 13 percent slower than the c6a.4xlarge, a notable achievement given that cloud instance has ten more CPU threads and four times as much memory.

The TPC-DS benchmark, which uses 24 tables and 99 more complex queries involving features like window functions, provided another perspective. At a standard scale factor (SF100), the MacBook Neo handled most queries effortlessly, with a median runtime of 1.63 seconds and a total completion time of 15.5 minutes. When pushed to a larger scale (SF300), the laptop’s memory constraints became apparent. While the median query runtime remained respectable at 6.90 seconds, the system had to use up to 80 GB of disk space for temporary data, a process known as “spilling.” One particularly complex query took 51 minutes to finish. Nonetheless, the hardware and software managed to complete the entire suite of tests in 79 minutes, demonstrating robust capability under pressure.

This isn’t the first time the core chip architecture has been stress-tested in an unconventional setting. When the iPhone 16 Pro launched, the same team ran the TPC-H benchmark with the device submerged in a bucket of dry ice at -50°C, where it completed the run in 478.2 seconds. These tests collectively highlight how efficient, integrated silicon designs can punch above their weight in data processing tasks, offering a compelling and portable alternative for certain analytical workloads.

(Source: 9to5Mac)