If you have old laptops, mini PCs, or discarded desktops gathering dust in your closet, you might be sitting on an untapped AI supercomputer.
As modern artificial intelligence models grow from 7 billion to 70 billion and 120 billion parameters, they require astronomical amounts of memory. Traditional tech advice tells you that old computers are useless for modern AI because they lack powerful modern graphics cards.
A growing movement of engineers and tech enthusiasts is proving that conventional wisdom completely wrong by turning old consumer electronics into distributed RAM pools.
The 120B Parameter Experiment: 6 Mismatched Devices
In a recent stress test that made waves across tech communities, an engineer connected six completely mismatched devices over standard home Wi-Fi and Ethernet:
- A Windows mini PC with an RTX 3060
- An aging 12 GB Acer laptop from 2019
- An Apple Mac mini M4
- A MacBook M3
- An old 2017 Intel MacBook Pro
- A Samsung Galaxy S24+ Android smartphone
Using RAMDeck‘s memory clustering technology, these six disparate machines pooled their memory together to load gpt-oss-120b—a colossal 60 GB AI model with 117 billion total parameters.
Surviving 40 Hours of Chaos
Running a massive frontier model on consumer hardware for two minutes is interesting, but running it for days across household Wi-Fi is an engineering feat. During a 40-hour continuous stress test:
- The cluster generated answers continuously at 1.37 tokens per second.
- When the Android phone dropped off Wi-Fi, the cluster automatically rebalanced the missing layers to other nodes without crashing.
- When a user closed the laptop lid on one of the MacBooks, the system seamlessly adapted and kept answering user prompts.
Instead of sending older electronics to landfills, memory pooling offers a second life for aging hardware, proving that you don’t need $10,000 server racks to experience frontier-tier AI.
Watch the live 40-hour stress test video and explore the cluster architecture at RAMDeck.com.
Leave a Reply