Hardware ·
My stepdad's 2012 PC is now my fleet's coordinator
A PC I built in 2012 sat in my parents' garage for years. Now it decides where my agents' tests run, and it has not frozen since I made it headless.
In 30 seconds
- A PC I built for my stepdad in 2012 sat in my parents' garage from about 2018. A new CMOS battery, a dust-out and Ubuntu brought it back.
- It is my fleet's coordinator. It picks the machine for every project's test jobs, builds the morning report and holds backups.
- On Oct 11 it froze for about 3 hours. That morning it went headless, got a hardware watchdog and started sending crash logs off the box.
- No freeze since. As of Oct 31, 2026, it has been up 2 weeks, 5 days and 22 hours.
Problem
In 2012 I built a PC for my stepdad: an Intel Core i7-2600 (4 cores, 8 threads) on an ASUS P8Z77-V LE Plus, with 16 GB of DDR3-1600. It worked great for years, until a hard drive started failing. Around 2018 he got a new machine, and the old tower went to my parents’ garage to gather dust.
Then I went all in on AI agents, and the fleet needed more hardware. I saw the tower sitting there and asked: can I just take it? They said yes, take it.
Could a desktop that old hold an always-on job that other machines lean on?
What I tried
| Attempt | What happened |
|---|---|
| A planned power-cut drill (Oct 9) | It came back with its settings wiped, stuck on an “F1” prompt. The CMOS battery was dead. |
| Running it like a normal desktop, graphics driver and all | It froze solid at 5:55 AM on Oct 11. The clock on its screen stopped and the network went silent until I power-cycled it at 9:00. |
| Looking for the cause | Nothing. The log stopped in the middle of normal work: low load, free memory, no error. The disks were healthy, and a 10 GB memory test came back clean. |
What worked
The revival. A bit of troubleshooting with my agents and a dust-out. Both old hard drives were identical twins: Seagate ST3000DM001s, the 3 TB model that Backblaze said about 32% of its own had failed by March 2015. One was dead, so it got unplugged. The other got wiped. Ubuntu 24.04 now runs from a Samsung 850 EVO SSD.
A new coin cell. After the failed drill I swapped in a fresh CR2032, re-applied the BIOS settings, and set it to power on after an outage and to wake on LAN. The repeat drill passed.
A real job. It is the fleet’s coordinator. It decides which machine runs each test job for every project, so the runs spread out. It also builds the daily morning report and holds backups. Several machines depend on it, so a frozen morning hurt.
Headless, the same morning. My main suspect is the old Sandy Bridge integrated-graphics driver, which has a known freeze with VT-d. It could also just be old hardware. Either way, I stopped the box loading that driver. The monitor now shows a plain text console, and the few tools that need a screen get a virtual one.
A watchdog, and logs that leave the box. The motherboard’s hardware watchdog is armed, so a freeze turns into a reboot in about a minute instead of three hours down. Crash logs now stream to another machine, so the next freeze has a better chance of leaving a trail.
Since then, quiet. As of Oct 31, 2026, it has been up 2 weeks, 5 days and 22 hours with no freeze, idling at a load of about 0.3 across its 8 threads.
Numbers
| Up since the Oct 11 reboot (as of Oct 31, 2026) | 2 weeks, 5 days, 22 h |
|---|---|
| Freezes since going headless | 0 |
| Load on 8 threads (Oct 31, 2026) | about 0.3 |
| Down in the Oct 11 freeze | about 3 h |
| Memory tested after the freeze | 10 GB, clean |
| Freeze to automatic reboot | about 1 min |
When it stops being true
I never proved the graphics driver did it. If it freezes again with that driver gone, I will blame the hardware, and the watchdog only buys time: it cannot fix a failing board or drive.
Headless suits a box nobody sits at. If you need a desktop on yours, this fix does not fit. And a board with no hardware watchdog leaves a freeze waiting for you to walk over and press the button.
Agent brief
Help me turn an old desktop into an always-on Linux box for light agent or CI work. "(general)" marks advice not from my setup.
First ask me:
- The CPU, board, memory and drives, and how long the machine sat unused.
- What job it should do and which machines will depend on it.
- Whether anyone needs a desktop on it, or it can run headless.
Constraints:
- No firmware updates or disk wipes without my go (general).
- Keep any old drive with data on it unplugged or untouched until I say otherwise.
- Give it light, always-on work, such as coordination, reports or backups, not heavy builds (general).
Steps:
- Have me dust it out and unplug any dead drive; check the remaining disks' health and run a memory test before trusting it (general).
- Install a current Ubuntu LTS on a solid-state drive.
- Do a planned power-cut drill. If it comes back with settings wiped or stuck on a firmware prompt, have me replace the CMOS coin cell.
- In the firmware, set power-on after an outage and wake on LAN, then repeat the drill until it passes.
- If nobody needs a desktop, run it headless: stop loading the integrated-graphics driver, keep a text console on the monitor, and give the few tools that need a screen a virtual one. Old Sandy Bridge graphics has a known freeze with VT-d.
- Arm the motherboard's hardware watchdog so a freeze reboots it in about a minute.
- Stream its logs to another machine so a freeze has a better chance of leaving a trail.
Pitfalls:
- A freeze can leave nothing in the local log: no error, low load, free memory. Logs shipped off the box are the best chance at a clue.
- Clean disks and a clean memory test do not rule out old hardware.
- The other machines that depend on it stall while it is down; watch it from one of them (general).
Done when:
- A power-cut drill brings it back with its settings intact, without me touching it.
- A test freeze triggers the watchdog and it reboots on its own (general).
- Its logs arrive on another machine, and it runs its job for a week with no freeze.