Shows / Big Iron Radio
Big Iron Radio
Single-narrator briefs and occasional deep dives on self-hosted, homelab, and local-AI topics - adapted from Big Iron guides. Produced with an AI-generated voice.
Episodes
- Big Iron Radio
Homelab Brief - Keeping SSDs Healthy: fstrim vs discard
TRIM keeps SSDs fast and evenly worn - the only question is when to send it. Why the weekly fstrim timer beats inline discard for almost every drive, plus the…
- Big Iron Radio
Homelab Brief - Does ZFS Really Need ECC RAM?
The 'ZFS requires ECC' debate, settled: ZFS doesn't require ECC and the scrub-of-death is a myth - but corrupt RAM still corrupts data on any filesystem. When…
- Big Iron Radio
Homelab Brief - Linux I/O Schedulers for HDD, SSD, and NVMe
Four Linux I/O schedulers, one rule: match the scheduler to the media. none for NVMe, mq-deadline for HDD and SATA/SAS SSD, and why bfq is great on a laptop…
- Big Iron Radio
RAID Showdown: Hardware RAID vs ZFS vs Btrfs vs mdadm
Speed is the wrong axis - the real question is whether your array can tell you when it's lying to you. Silent bit-rot, why checksums change everything, and a…
- Big Iron Radio
Homelab Brief - Training LoRAs Locally with Kohya-ss
Train your own image-model LoRAs locally with kohya-ss - and why the dataset is 80% of the result. Produced with an AI-generated voice; adapted from a Big Iron…
- Big Iron Radio
Homelab Brief - Offloading Immich Machine Learning to a GPU Box
Offload Immich's CLIP and face-recognition ML onto a dedicated GPU box - one env var, officially supported. Produced with an AI-generated voice; adapted from a…
- Big Iron Radio
Homelab Brief - GPU Memory Fragmentation and OOM During Inference
The CUDA out-of-memory error that fires while nvidia-smi shows plenty free - diagnosing and fixing fragmentation. Produced with an AI-generated voice; adapted…
- Big Iron Radio
Homelab Brief - eGPUs on Linux: Thunderbolt, OCuLink, and the Bandwidth Tax
Running an external GPU on Linux, and why the 'bandwidth tax' matters less than the benchmarks suggest. Produced with an AI-generated voice; adapted from a Big…
- Big Iron Radio
The 2026 Homelab Workstation That Also Runs LLMs
One machine that's your workstation and your LLM box - the real compromise lives in the chassis, PSU, and airflow, not the compute parts. Produced with an…
- Big Iron Radio
Homelab Brief - Prompt-Injection Defenses: Input and Output Guards
Realistic prompt-injection defense for self-hosted LLM apps: contain the blast radius, because you can't prevent it. Produced with an AI-generated voice…
- Big Iron Radio
Homelab Brief - Hardening MCP Servers: Auth and Transport
Hardening MCP servers once they become network services instead of local subprocesses. Produced with an AI-generated voice; adapted from a Big Iron guide. Full…
- Big Iron Radio
Homelab Brief - An MCP Gateway: One Endpoint, Many Servers
When to put a gateway in front of your MCP servers - one endpoint for an LLM to reach many tools. Produced with an AI-generated voice; adapted from a Big Iron…
- Big Iron Radio
Homelab Brief - Safe Agent Writes: Approval Gates and Action Ledgers
Let an agent make changes safely with the write-path pattern: propose, review, apply. Produced with an AI-generated voice; adapted from a Big Iron guide. Full…
- Big Iron Radio
Exposing Your Homelab to an LLM: The Read-Only MCP Server Pattern
Let an LLM answer 'how's my homelab doing?' safely with a read-only MCP server - where the security boundary is which tools exist, not what their docstrings…
- Big Iron Radio
Homelab Brief - Obsidian as Long-Term Memory for an LLM
Give an LLM agent persistent, human-readable memory: a markdown vault over MCP that you can open, read, and correct. Produced with an AI-generated voice…
- Big Iron Radio
Homelab Brief - Embedding Throughput: CPU vs iGPU vs dGPU
How fast you can build RAG embeddings on CPU, iGPU, and dGPU - and which one you actually need. Produced with an AI-generated voice; adapted from a Big Iron…
- Big Iron Radio
Homelab Brief - Auto-Tagging Your Photos with CLIP and pgvector
Auto-tag a photo archive with CLIP in a pipeline that outlives whatever app you use today. Produced with an AI-generated voice; adapted from a Big Iron guide…
- Big Iron Radio
Homelab Brief - Batch Photo Captioning with Vision LLMs
Caption a photo library with a local vision LLM, written to metadata that survives app migrations. Produced with an AI-generated voice; adapted from a Big Iron…
- Big Iron Radio
Filesystem-Only RAG: When You Don't Need a Vector Database
Most homelab RAG doesn't need a vector database - build the whole thing in an afternoon with markdown on disk, and spend your effort where it actually moves…
- Big Iron Radio
Homelab Brief - One GPU for Immich, Frigate, and Whisper
Fitting Immich, Frigate, and Whisper onto one GPU - where the VRAM goes and how to stop them fighting. Produced with an AI-generated voice; adapted from a Big…
- Big Iron Radio
Homelab Brief - Semantic Camera Search with Frigate and Ollama
Add semantic search and local AI descriptions to Frigate with Ollama, keeping footage on your own network. Produced with an AI-generated voice; adapted from a…
- Big Iron Radio
Homelab Brief - Frigate Detectors: Coral vs OpenVINO vs TensorRT
Picking a Frigate object detector in 2026, after the Coral's value collapsed - Coral vs OpenVINO vs TensorRT. Produced with an AI-generated voice; adapted from…
- Big Iron Radio
Homelab Brief - TTS Announcements Across the House with Snapcast
Inject doorbell chimes and TTS announcements into a whole-house Snapcast setup, which has no native duck-and-resume. Produced with an AI-generated voice…
- Big Iron Radio
Building a Fully Local Voice Assistant: Whisper, Piper, and Wyoming
Build a fully local voice assistant with Wyoming, Whisper, Piper, and openWakeWord - and the one sizing decision that makes or breaks it. Produced with an…
- Big Iron Radio
Homelab Brief - Local Agents vs Cloud APIs: The Break-Even Math
The honest break-even math for a 24/7 AI agent: a dedicated GPU's fixed plus idle cost versus cloud pricing that's zero when idle. Produced with an…
- Big Iron Radio
Homelab Brief - A Local LLM Voice Assistant in Home Assistant
Wire a local LLM into Home Assistant the right way: let the built-in intent matcher handle most commands, and route only the fuzzy remainder to the model…
- Big Iron Radio
Homelab Brief - VRAM Budget: Which LLM Actually Fits Your GPU
The back-of-the-envelope VRAM math for local LLMs: weights, the KV cache everyone forgets, and overhead. Produced with an AI-generated voice; adapted from a…
- Big Iron Radio
Homelab Brief - Managing the Context Window in Long Agent Loops
Keeping a long-running agent from drowning in its own context - quality degrades long before the window overflows. Produced with an AI-generated voice; adapted…
- Big Iron Radio
Homelab AI in 2026: What's Real and What's a Gimmick
A filter, not a hype piece: which 2026 homelab-AI projects actually survive, which get quietly turned off, and the five-question test to tell them apart before…
- Big Iron Radio
Welcome to Big Iron Radio
Short, practical audio for people who run their own hardware - homelab, self-hosted, IT, and local-AI decisions, one honest verdict at a time. Quick briefs…