How noisy and power-hungry are dedicated AI rigs at home?
Much less expensive to run than most people expect, though a high-end PC rig will temporarily sound like a gaming tower under full load.
Chatting with an AI happens in short 5-to-15-second bursts rather than continuous hours of heavy compute, so your hardware spends roughly 95% of its time sitting at idle. Noise, heat, and electricity draw still vary widely depending on the platform you choose:
| Rig Type | Idle Power | Peak Load Power | Est. Monthly Electricity (2 hrs active/day) | Noise Level |
|---|---|---|---|---|
| Apple Mac Studio / MacBook Pro | 10 – 15 W | 70 – 170 W | ~$2 – $4 / mo | Whisper-quiet (barely audible) |
| Mid-Range PC (e.g., RTX 4060 Ti / 5070 Ti) | 35 – 50 W | 250 – 380 W | ~$8 – $12 / mo | Moderate (gentle fan hum) |
| Enthusiast PC (e.g., RTX 3090 / 4090) | 45 – 70 W | 450 – 600 W | ~$12 – $18 / mo | Audible (whoosh of GPU fans during replies) |
| Dual-GPU Rig (e.g., Dual RTX 3090s) | 70 – 100 W | 750 – 850 W | ~$20 – $30 / mo | Loud + Warm (acts like a small space heater) |
Note: Monthly electricity estimates assume ~22 hours/day at idle and ~2 hours/day of active AI generation at average residential electricity rates.
Three things every home builder should know
- Chatting is bursty; image and video generation are continuous. Typing prompts to a local LLM spikes your GPU for a few seconds at a time while tokens stream, costing pennies a day in power. Generating AI videos or fine-tuning models pegs your GPU at 100% load for hours—warming up a small room and spinning case and GPU fans at full speed.
- Watch out for blower-style and passive server cards. If you buy used workstation GPUs, check the cooler design before ordering. Cards like the NVIDIA RTX A6000 use a single-fan "blower" cooler built for loud server racks and sound like a hairdryer under load, while older datacenter cards like the Tesla P40 ship without any onboard fan and require high-static-pressure rack airflow or loud aftermarket fan shrouds. Keep server-style hardware in a basement, garage, or closet rather than next to your desk.
- Cut power and noise by 20–30% with almost zero speed loss. Text generation is bottlenecked mostly by memory bandwidth rather than peak GPU core voltage. Capping your GPU's power limit by 20–30% using free tools like
MSI Afterburnerornvidia-smidrops fan noise and heat output dramatically while costing less than 5% of your token generation speed.
If your rig sits right beside your monitor in a quiet bedroom or office, an Apple Silicon Mac or a power-limited single-GPU PC keeps things peaceful. If you want a high-wattage dual-GPU rig, plan to put it under your desk—or run it as a headless server in another room and connect over your home network.
Looking for verified hardware rigs?
Check our real-time £/GB VRAM index to find systems optimized for local model inference.