In a fascinating benchmark conducted by DuckDB researcher Gábor Szárnyas, the 512 GB MacBook Neo was pitted against various high-end cloud servers in heavy database workloads. Published on Boing Boing under the title “Big Data on the Cheapest MacBook,” the results highlight the surprising raw power of Apple’s entry-level laptop. Szárnyas utilized two primary industry-standard benchmarks for this analysis: ClickBench and TPC-DS.
MacBook Neo vs. Cloud Giants: The Performance Breakdown
The ClickBench test focuses on aggregation and filtering across a single wide table containing 100 million rows, utilizing 14 GB in Parquet format and 75 GB in CSV. The TPC-DS test is more complex, involving 24 tables and 99 queries with window functions. The MacBook Neo competed against the c6a.4xlarge (16 AMD EPYC cores, 32 GB RAM) and the massive c8g.metal-48xl (192 Graviton4 cores, 384 GB RAM).
Cold vs. Hot Cache Results
- Cold Run: In the “cold” phase where caches are empty, the MacBook Neo outperformed its rivals by up to 2.8x, completing all queries in under a minute. This victory is attributed to the MacBook’s local NVMe SSD compared to the cloud servers’ network-attached storage.
- Hot Run: In the “hot” phase with populated caches, the massive c8g.metal-48xl led with 4.35 seconds. The MacBook Neo finished last at 54.27 seconds, but impressively, it remained only 13% slower in total time than the mid-tier cloud server, despite having 10 fewer processing threads and 4x less RAM.
Pushing the Limits: TPC-DS SF100 and SF300
During the TPC-DS SF100 stage, the laptop finished all tests in a total of 15.5 minutes with an average query latency of 1.63 seconds. The more demanding SF300 stage pushed the memory limits, requiring 80 GB of swap space on the disk. Although query 67 took 51 minutes to complete, the seamless hardware-software integration allowed the device to successfully finish the entire suite in 79 minutes.
A History of Extreme Testing
Interestingly, this team is no stranger to pushing Apple Silicon to its limits. Previously, they tested the A19 Pro chip by placing an iPhone 16 Pro in a bucket of dry ice at -50°C to run the TPC-H benchmark, which it completed in 478.2 seconds. These tests continue to prove that Apple’s “entry-level” hardware can handle workloads typically reserved for server-grade infrastructure.
What are your thoughts on the performance of Apple’s latest silicon? Would you consider such a portable model for heavy data engineering tasks,
