vllm

4 articles

LLM Inference Metrics: TTFT, ITL, TPS and Goodput

A 2026 worksheet for comparing LLM inference benchmarks: define TTFT, ITL, per-user and aggregate TPS, normalize workload and timing boundaries, and reject unsupported goodput claims.

9/28/2026• 24 min read

llm inference metricsttftinter-token latency

Triton 26.08 Upgrade Runbook: CUDA 13.4, Canary and Rollback Testing

A 2026 upgrade runbook for Triton Inference Server 26.08 covering CUDA 13.4.1 compatibility, container image selection, backend validation, canary testing, and rollback criteria.

9/24/2026• 24 min read

triton 26.08 upgrade guidetriton upgrade runbookinference server upgrades

Disaggregated Prefill/Decode Private Fleet Break-Even

A 2026 break-even model for private disaggregated prefill/decode fleets, covering SLO goodput, KV transfer, idle pool-hours, HBM opportunity cost, and conditional routing.

9/22/2026• 22 min read

disaggregated prefill decodeprefill decode disaggregationllm inference

NVIDIA Dynamo 1.4.2 Enterprise Support Checklist

A 2026 enterprise readiness guide to Dynamo 1.4.2 artifacts, backend and CUDA pins, the NIXL loader fix, failure coverage, pilot metrics, and contract questions.

9/19/2026• 21 min read

nvidia dynamo 1.4.2 enterprise supportnvidia dynamoenterprise inference