Skip to content
Tech News
← Back to articles

Benchmark finds local NVMe outperforms gp3 EBS on 3.3B-row Postgres workload

read original get Samsung 990 Pro NVMe SSD → more articles
GoKawiil Brief

At POSETTE 2026, Sai Srirampur presented a benchmark comparing a 3.3-billion-row PostgreSQL workload running on baseline gp3 EBS storage versus local NVMe storage. The talk argued that common scaling symptoms—slow ingestion, inconsistent read latency, autovacuum falling behind, and slower checkpoints—often trace back to storage performance rather than Postgres itself, and proposed architecture patterns for running Postgres on local NVMe in production.

Why It Matters

GoKawiil's interpretation of the reporting above, not reported fact.

The findings suggest that teams troubleshooting Postgres performance issues at scale may be addressing symptoms in the database layer when the underlying bottleneck is actually disk I/O, according to the presenter. If local NVMe meaningfully reduces these problems, it could reshape how production Postgres deployments are architected for storage, though adopting local NVMe introduces its own durability and failover tradeoffs not detailed here.

Key Takeaways
Worth a Look

Samsung 990 Pro NVMe SSD — If you're testing local NVMe storage performance for database workloads like the Postgres benchmark described here, a fast NVMe drive like the Samsung 990 Pro is a great way to experiment with low-latency storage at home or in a dev rig. It's a solid pick for anyone wanting to see firsthand how storage speed impacts I/O-heavy workloads like PostgreSQL ingestion and checkpointing.

See Samsung 990 Pro NVMe SSD on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.

Source: clickhouse.com, 2026-10-12

Published there as: “Postgres isn't slow, your storage is”

Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.