Perf engineer praises memory-overcommit evals, suggests exposing stats to fleet scheduler
sloppenheimer · x · 2026-09-24
The author highlights the evaluations section of a deployment writeup, noting that memory-overcommit behavior over time looks clean and largely contention-free. He suggests exposing these stats to the scheduler could unlock better fleet-level distribution wins at scales above 160 nodes. He also recalls past struggles with NVIDIA MIG in vGPU/microVM setups, says the FnCall approach sidesteps them, and wonders whether an nvidia-smi-based external reset exists for when an agent trashes a GPU.
Related event: Users Critique Systems Paper's Evaluation, Missing gVisor Benchmarks(2 posts)→
More from coding & agent
- Running Qwen locally with Hermes agent in 10 minutes on an old gaming laptop — markjeffrey · 2026-09-24
- The Jev + Claude Code combo: a 100ms, near-free decision brain that blocks destructive AI commands — PrajwalTomar_ · 2026-09-24
- Resend ships Stripe Projects integration as the most-requested email provider — jeff_weinstein · 2026-09-24
- Anthropic cut Opus 5.5 prices, then broke four things your agent depends on — rseroter · 2026-09-24
- treg.to skill taps 60 data providers for agent lead enrichment: 20 verified emails for $0.58 — rohanpaul_ai · 2026-09-24
- Netflix hiring PhD intern to build LLM AutoResearch agents for recommendations — tw_killian · 2026-09-24