Insights and hands-on guides on frontend, backend, databases, IT infrastructure, and AI solutions.
Our Servers Catch Burglars Every Day — TREEWALL Field Test Report
NVMe Storage Benchmark: Optane to HDD — 20 Devices Compared with fio
Local AI Video Generation with ComfyUI + Wan 2.2 — An RTX PRO 6000 96GB Build Log
Structuring Unstructured Documents with an LLM — Measured Accuracy on a 19-Item Menu
Guarding the Front Door of Your Network — TREEWALL, an AI Security Gateway
A local 14B model turns menu documents into database records: 100% on prices, 94.7% on names. The dangerous errors aren't the obviously wrong values — they're the plausibly wrong ones.
When a healthy model file outputs nothing but nonsense, where do you start? A three-step trail to the real root cause.
We tested LG AI Research's EXAONE 4.5-33B on RTX PRO 6000 on launch day. FP8 failed, custom fork needed, BF16 at 22 TPS, 100% safety. Full 12-scenario 85-question benchmark results.
Gemma 4 benchmark on RTX PRO 6000 Blackwell — 8 setup attempts, 3 model comparison (Qwen3-32B, Gemma4-31B, Gemma4-26B MoE), and the surprising MoE speed advantage.
NVIDIA driver 590→595 upgrade benchmarks on RTX PRO 6000 Blackwell — SGLang inference speed, CUDA compute, P-State changes, and operational warnings.
Qwen3-32B is 2x slower than 14B. We tested both with 60 identical questions to measure exactly how much quality you gain for the speed trade-off.
AWQ INT4 delivers 1.88x at 4B scaling to 2.94x at 32B. MoE Qwen3-30B-A3B runs faster than Dense 14B-AWQ. 16 models benchmarked with full data on RTX PRO 6000.
The RTX 5060 Ti runs 8B models at 76 tok/s and handles 30 concurrent users with zero errors. Here is the complete benchmark at $450.
Running 7 tenants on one GPU with 5 LoRA adapters, we executed 15 isolation tests to verify zero cross-tenant data leakage.
vLLM looks faster at median latency. But at P95, SGLang wins by 6.5x. With 3x throughput and zero errors at 200 users, we explain why P95 matters more than median for production LLM serving.
Build a CCTV Recording Server Without an NVR — RTSP + ffmpeg + systemd, 180-Day Retention
Our MoE Model Started Spewing Garbage — What Auditing All 18,432 Expert Weights Revealed
What We Learned Building and Auditing 990 AI-Generated English Learning Scenes
BGE-M3 vs mE5-Large — Comparing Korean RAG Search Quality Head to Head
Recovering winget and Windows Terminal on a Windows 11 Box Where the Microsoft Store Won't Open
