Oriel — AI & technology worth noticing, every day.

It's the Memory Latency, Stupid

Web page · https://pivotlake.io/blog/its-the-memory-latency-stupid

First seen by Oriel on .

AI summary

Pivot’s GROUP BY benchmark shows that reducing DRAM accesses through radix partitioning substantially improves single-core performance, despite performing more CPU operations.

  • A naive HashMap aggregation over 99m rows and approximately 17m unique UserID values took 3.6 seconds on its second run.
  • The naive implementation spent 50.1% of CPU slots stalled on memory, retired operations during 16.1%, and pulled 4.5GB from DRAM.
  • Radix partitioning hashes keys into 4096 partitions, then aggregates each smaller partition with a separate HashTable that is more cache-friendly.
  • The optimized version reduced second-run time to 2.7 seconds, memory stalls to 23.5%, and DRAM traffic to 1.5GB.
  • Pivot’s further implementation completed the query in 1.848 seconds on one core and pulled 954MB from DRAM.

#databases #cpu caches #radix partitioning #pivot

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Points

Score reported by the source.

  • Hacker News: 270, latest observation ; +265 over observed period since

Comments

Comment count reported by the source.

  • Hacker News: 101, latest observation ; +101 over observed period since

Aliases

SourceExternal IDURL
Hacker News50318720https://pivotlake.io/blog/its-the-memory-latency-stupid/ Read discussion