Engineering blog.

A record of what we measured and decided in the field. Published biweekly — for HPC/AI operators.

Recent posts.

• Ops2026-04-22·9 min

  GPU Utilization from 30% to 78% — What We Did Over Six Months

  How we more than doubled GPU utilization on a university AI cluster by redesigning queue policy and giving users visibility.

  — Engineering Team

• Tech2026-04-08·11 min

  NCCL all-reduce Efficiency 60% → 96% — Rail-Optimized and Adaptive Routing

  On a 64-node H100 PoC, NCCL all-reduce delivered only 60% of spec. How we drove it to 96% through topology and routing policy.

  — AI Infrastructure Team

• Product2026-03-25·8 min

  Luxe Vantage Billing Design — How to Build Automatic Per-Department Chargeback

  Can you get cloud-like billing on an on-premises GPU cluster? Slurm sacct + Kubernetes metrics + per-department rate policy.

  — Product Team

• Industry2026-03-11·7 min

  The RFP Price-Weighting Trap — Why 30% Is the Right Number

  What happens when the price weighting in a public-sector RFP exceeds 50%. Designing a scorecard that yields infrastructure built to survive five years of operations.

  — Consulting Team

• Tech2026-02-20·6 min

  Six Ways We Cut LCP by 0.4s on a CJK Site — Starting with Pretendard Self-Hosting

  The LCP of a CJK-language B2B site runs on average 0.5s longer than an English site. Here is how we cut it.

  — Frontend Team

• Ops2026-02-04·7 min

  Zero-Touch Boot — Redeploying the OS on 200 Nodes in 12 Minutes

  Every major GPU driver update means reprovisioning 200 nodes. How Luxe Orbit cut it to 12 minutes.

  — Operations Team

• Tech2026-01-18·9 min

  IB vs. RoCEv2 Decision Framework — Resolving a 28% Cost Gap on Three Axes

  The workloads where InfiniBand is the answer and where RoCEv2 is. Decide on three axes: cost, operational experience, and NCCL efficiency.

  — Network Team