In 2024 I built a Pi Frigate NVR with Axzez's Interceptor 1U Case , and installed it in my 19" rack. Using a Coral TPU for object detection, it's been dutifully surveilling my property—on my terms (100% local, no cloud integration or account required). I've wanted to downsize the setup while keeping cheap large hard drives 1 , and an AI accelerator.
Apparently Apple thinks nobody with a modern Mac uses spinning rust (hard drives with platters) anymore. I plugged in a hard drive from an old iMac into my Mac Studio using my Sabrent USB to SATA Hard Drive enclosure, and opened up Disk Utility, clicked on the top-level disk in the sidebar, and clicked 'Erase'. Lo and behold, there's no 'Security Options' button on there, as there had been since—I believe—the very first version of Disk Utility in Mac OS X!
I run Frigate to record security cameras and detect people, cars, and animals when in view. My current Frigate server runs on a Raspberry Pi CM4 and a Coral TPU plugged in via USB. Raspberry Pi offers multiple AI HAT+'s for the Raspberry Pi 5 with built-in Hailo-8 or Hailo-8L AI coprocessors, and they're useful for low-power inference (like for image object detection) on the Pi. Hailo coprocessors can be used with other SBCs and computers too, if you buy an M.2 version .
Over the weekend Ars Technica retracted an article because the AI a writer used hallucinated quotes from an open source library maintainer. The irony here is the maintainer in question, Scott Shambaugh, was harassed by someone's AI agent over not merging it's AI slop code. It's likely the bot was running through someone's local 'agentic AI' instance (likely using OpenClaw). The guy who built OpenClaw was just hired by OpenAI to "work on bringing agents to everyone." You'll have to forgive me if I'm not enthusastic about that.
When I saw Jensen Huang introduce the Reachy Mini at CES , I thought it was a gimmick. His keynote showed this little robot responding to human input, turning its head to look at a TODO list on the wall, sending emails, and turning drawings into architectural renderings with motion. HuggingFace and Pollen robotics sent me a Reachy Mini to test, and, well, at least if you're looking to replicate that setup in the keynote, it's not, as Jensen put it, "utterly trivial now."
In October, my Dad and I got to go behind the scenes at two St. Louis Blues (NHL hockey) games, and observe the massive team effort involved in putting together a modern digital sports broadcast. I wanted to explore the timing and digital side of a modern SMPTE 2110 mobile unit, and my Dad has been involved in studio and live broadcast for decades, so he enjoyed the experience as the engineer not on duty!
Ever since the Raspberry Pi Compute Module 5 was introduced, I wondered why nobody built a decent laptop chassis around it. You could swap out a low spec CM5 for a higher spec, and get an instant computer upgrade. Or, assuming a CM6 comes out someday in the same form factor, the laptop chassis could get an entirely new life with that upgrade.
Recently this post from @Merocle caught my eye: I'm fixing my iFixit soldering station. I haven't used it for a long time and the battery has gone overdischarge. I hope it will come back to life. Unfortunately, there are no replacements available for sale at the moment. Devices with built-in rechargeable batteries have been bugging me a lot lately. It's convenient to have a device you can take with you and use anywhere. And with modern Li-ion cells, battery life is remarkable.
A few weeks ago, I started hearing a slight crackle at the loudest parts of whenever sound was playing through my PreSonus Eris E3.5 speakers . It was very faint, but quite annoying, especially when editing my YouTube videos. For a few days I thought it could be a hearing problem (at this point in my life, every year brings a new health adventure...), but after testing my wired headphones and another small computer speaker on the same output, I determined the problem was, indeed, coming from the PreSonus speakers.
After 16 years on the LAMP stack, I finished migrating this website from Drupal to Hugo a few weeks ago. What's old is new, as this blog was originally built with Thingamablog , a Java-based Static Site Generator (SSG) I ran on my Mac to generate HTML and FTP it up to my first webserver (over 20 years ago!). The main reason I moved from an SSG to Drupal was to add comments . I wanted my blog to have the same level of interactivity I had pre-Thingamablog, when I was (briefly) on Xanga.com.
Today Raspberry Pi launched their new $130 AI HAT+ 2 which includes a Hailo 10H and 8 GB of LPDDR4X RAM . With that, the Hailo 10H is capable of running LLMs entirely standalone, freeing the Pi's CPU and system RAM for other tasks. The chip runs at a maximum of 3W, with 40 TOPS of INT8 NPU inference performance in addition to the equivalent 26 TOPS INT4 machine vision performance on the earlier AI HAT with Hailo 8.
I wanted to have the most accurate timepiece possible mounted in my mini rack. Therefore I built this: This is a GPS-based clock running on a Raspberry Pi Pico in a custom 1U 10" rack faceplate. The clock displays time based on a GPS input, and will not display time until a GPS timing lock has been acquired. For full details on designing and building this clock, see: When you turn on the Pico, the display reads Upon 3D fix, you get a time on the clock, and the colon starts blinking If the 3D fix is lost, the colon goes solid When the 3D fix is regained, the colon starts blinking again
For the past decade, I've used Mailhog for local email debugging. Besides working on web applications that deal with email, I've long used email as the primary notification system for comments on the blog. I built an Ansible role for Mailhog , and it was one of the main features of Drupal VM , a popular local development environment for Drupal I sunset 3 years ago. Unfortunately, barring any future updates from the maintainers, it seems like Mailhog has not been maintained for four years now. It still works , but something as complex as an email debugging environment needs ongoing maintenance to stay relevant.
Almost a year ago, I found that N100 Mini PCs were cheaper than a decked-out Raspberry Pi 5 . So comparing systems with: Back in March last year, a GMKtec Mini PC was $159, and a similar-spec Pi 5 was $208. Today? The same GMKtec Mini PC is $246.99, and the same Pi 5 is $246.95: Today, because of the wonderful RAM shortages 1 , the Mini PC is the same price as a fully kitted-out Raspberry Pi 5. 16GB of RAM 512GB NVMe SSD Including case, cooler, and power adapter
Dell sent me two of their GB10 mini workstations to test: In this blog post, I'll cover the base system, just one of the two nodes. Cluster testing is ongoing, and I'll cover things like AI model training and networking more in depth next year, likely with comparisons to the Framework Desktop cluster and Mac Studio cluster I've also been testing. But many of the same caveats of the DGX Spark (namely, price to performance is not great if you just want to run LLMs on a small desktop) apply to Dell's GB10 box as well.
If you were 5 microseconds late today, blame it on NIST. Their facility in Boulder Colorado just had its power cut for multiple days. After a backup generator failed, their main ensemble clock lost track of UTC, or Universal Time Coordinated. But even if you used the NTP timing servers they run , they were never off by more than 5 microseconds. 5 μs might seem insignificant. But it is significant for scientists and universities who rely on NIST's more specialized timing signals . But no, you don't need to panic. And yes, they have it under control now. But I thought I'd go over what happened, what it means, and what we can learn from NIST's near-outage. This blog post is a lightly-edited transcript of my most recent YouTube video:
Ever since I got AMD , Intel , and Nvidia graphics cards to run on a Raspberry Pi, I had a nagging question: What's the point? The Raspberry Pi only has 1 lane of PCIe Gen 3 bandwidth available for a connection to an eGPU. That's not much. Especially considering a modern desktop has at least one slot with 16 lanes of PCIe Gen 5 bandwidth. That's 8 GT/s versus 512 GT/s. Not a fair fight.
Apple gave me access to this Mac Studio cluster to test RDMA over Thunderbolt, a new feature in macOS 26.2 . The easiest way to test it is with Exo 1.0 , an open source private AI clustering tool. RDMA lets the Macs all act like they have one giant pool of RAM, which speeds up things like massive AI models.
This little postage stamp is actually a full Raspberry Pi Zero 2, complete with eMMC storage and WiFi. But you can't get one. Well, not unless you buy the CM0NANO development board from EDAtec , or you live in China. This little guy doesn't have an HDMI port, Ethernet, or even USB. It's a special version of the 'Compute Module' line of boards. Little Raspberry Pi 'System on Modules' (SoMs), they're called. Compute Modules are entire Linux computers about the size of a regular desktop CPU that you 'plug in' to another board, to give it life.
Now that Nvidia GPUs run on the Raspberry Pi , I've been putting all the ones I own through their paces. Many people have an older Nvidia card (like a 3060) laying around from an upgrade. So could a Pi be suitable for GPU-accelerated video transcoding, either standalone for conversion, or running something like Jellyfin for video library management and streaming? That's what I set out to do, and the first step, besides getting the drivers and CUDA going (see blog post linked above), was to find a way to get a repeatable benchmark going.
AI
Tutorial
Hardware
Security
Rust
Programming
Open Source
UI
HTML
Java
Web Development
DevOps
Business
Performance