LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience
abhijithneil · x · 2026-09-11
A segment of a thread on LLM serving metrics:
- TPOT (time per output token): the streaming speed after the first token. If too slow, users start reading ahead of the model, which feels terrible.
- Uptime and availability: a model that's fast 90% of the time but down the other 10% will still hurt the user experience.
Related event: Survey highlights TPOT and SLA differences in LLM inference services(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
- PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11
- Can a 7900 XTX 24GB run Qwen locally? Reddit seeks ROCm tok/s benchmarks — thenomadexplorerlife · 2026-09-11