Funny how a fanless 8-core laptop can chew through TPCH once you pair a columnar layout with vectorized execution; at that point the bottleneck is memory bandwidth, and Apple s unified LPDDR ends up outpacing plenty of noisy rack servers.
On paper the Neo's 120 GB/s LPDDR isn't huge, but when the whole working set lives in that pool the predictable latency lets the vector engine stretch its legs. I still wouldn't schedule nightly ETL on a laptop, but it's a good reminder that a lot of so-called big data fits in RAM just fine.
Funny how a fanless 8-core laptop can chew through TPCH once you pair a columnar layout with vectorized execution; at that point the bottleneck is memory bandwidth, and Apple s unified LPDDR ends up outpacing plenty of noisy rack servers.
On paper the Neo's 120 GB/s LPDDR isn't huge, but when the whole working set lives in that pool the predictable latency lets the vector engine stretch its legs. I still wouldn't schedule nightly ETL on a laptop, but it's a good reminder that a lot of so-called big data fits in RAM just fine.