What Happens When You Run Two GPU Batch Jobs on One Circuit

I had two machines running Whisper transcription at the same time. My Mac M2 Max was using Metal to process course transcripts. My PC with a GTX 1080 Ti was using CUDA to do the same thing on a different set of files. Both at full GPU load.

Both plugged into the same electrical circuit.

What happened next was predictable in retrospect. But at the time, I was confused why my PC kept shutting down mid-batch.

The Symptoms

The first time it happened, I thought it was a bad file. The GTX 1080 Ti was chugging through a course on copywriting and just... died. Black screen. No warning.

I rebooted, restarted the batch, and it crashed again on a different file. Then a third time.

I blamed the Whisper installation. Maybe the CUDA DLL workaround was flaky. Maybe the GPU was overheating. I spent an hour debugging the wrong problem.

Then it happened on the Mac too. Not a crash, but everything went dark for a second. The monitors, the desk lamp, the VoIP phone. All of it. And then it came back.

That's not a GPU driver issue. That's a circuit breaker.

The Power Draw

Once I figured out what was happening, I did the math on what was actually on that circuit.

Mac M2 Max under Metal load: 50-70W PC (i7-3970K + GTX 1080 Ti under CUDA load): 350-400W Second MacBook + monitor: 60W UniFi Dream Machine Pro: 15-20W UniFi 24-port PoE switch + 5 Raspberry Pis: 60-80W HDHomeRun, MoCA adapters, Hue hub, VoIP phone: 20W Two laser printers (idle): 20-30W Mini fridge: 100-150W (when compressor kicks on)

That's roughly 700-850W steady state. Most home circuits are 15 amps, which is 1,800W. On paper, we're fine.

But circuits don't trip at their rated capacity. They trip based on the breaker's time-current curve. A steady 800W load with a sudden spike from the mini fridge compressor cycling on at the same moment the GPU pushes harder during a complex transcription pass? That spike can push you over the threshold momentarily.

The Real Culprit

The mini fridge.

A mini fridge compressor doesn't run continuously. It cycles on and off, typically pulling 3-5x its rated running power for the first second or two when it kicks on. That inrush current is the spike that pushed the circuit over the edge at exactly the wrong moment.

Everything else on the circuit was fine. The steady-state load was well within limits. But you can't predict when the compressor will cycle, and you can't control when a GPU will spike during inference.

The combination was random but frequent enough to make both batch jobs unreliable.

The Fix

The solution was simple. I moved the mini fridge to a different circuit.

After that, both machines ran their batch jobs for days without a single crash. The Mac processed 457 files. The PC processed its queue. No reboots, no lost progress, no mystery shutdowns.

The Lesson for Solopreneurs

This is the kind of problem nobody warns you about when you start running a business from home. There's no course about electrical load balancing. No LinkedIn influencer posting about circuit breaker math.

When you're one person running your own infrastructure, you become responsible for everything. Not just the code and the content, but the physical environment it runs in. Power, cooling, networking, storage. It's all on you.

One of the courses I went through talked about how solopreneurs hit walls when they try to scale alone. The usual advice is about hiring or delegating. But sometimes the wall isn't about people. It's about physics.

I was trying to do two heavy compute jobs simultaneously on consumer-grade electrical wiring in a home office. The constraint wasn't my skill or my tools. It was the building I was sitting in.

The fix cost me nothing. Unplugging a mini fridge and walking it to the kitchen. But diagnosing the problem took hours because I was looking in the wrong place. I assumed it was a software or hardware failure when it was a power infrastructure issue all along.

If you're running serious hardware at home, map your circuits. Know what's on each breaker. It's not glamorous work, but it saves hours of debugging mysterious crashes that have nothing to do with your code.

And if you're running two GPU batch jobs at the same time? Put the mini fridge somewhere else.