Random Engram Reads Kill SSDs: NVLink-Backed CPU Memory and Redis Eyed for Sparse Memory Lookup
bookwormengr · x · 2026-09-11
A technical X thread on serving sparse lookup-based memory (Engram) at scale: the author notes Hopper lacks NVL72 racks (the diagram was illustrative), and argues Engram's highly random reads make SSDs a poor fit for server-class use. Reading the Engram table from CPU memory over NVLink lets any of 72 GPUs running the model backbone access a single copy with low latency; an in-memory store like Redis could push lookup latency to single-digit milliseconds.
Related event: Engram's random reads strain SSDs, engineers debate in-memory solutions(2 posts)→
More from Infra
- Engram's random reads don't suit SSDs; CPU-memory over NVLink could serve all 72 GPUs — bookwormengr · 2026-09-11
- 80% of the DIY LLM inference hype posters have already quit — it's brutally hard systems work — abhijithneil · 2026-09-11
- Hugging Face's Ultra Scale Playbook: a free book on training LLMs on GPU clusters — mdancho84 · 2026-09-11
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11
- LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience — abhijithneil · 2026-09-11
- PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11