One AI Model Now Runs on Five Gadgets at Once
This could let you run powerful AI on devices you already own.
A tech tinkerer chained five of their own devices — two gaming PCs (an RTX 2060 Super and an RX 9060 XT), an M4 Pro MacBook, and two iPhone 17 Pro Maxes — into one system to run Qwen 3.8 Flash Next, a 48-layer mixture-of-experts model with 512 experts per layer that's far bigger than any single machine they own. The trick is pipelined speculation: a draft head on PC 1 guesses the next few tokens, the whole chain checks them in one pass, and PC 1 is already drafting the next batch while the current one travels through the Mac, the phones, and PC 2.
Each step lands about 2.2 tokens on average. Measured at the server with what an agent actually sends, it writes at 42 tok/s at 16k context, 39–41 tok/s at 32k, and 37–39 tok/s at 60k, while reading prompts at 728–744 tok/s, 844–852 tok/s, and 856–895 tok/s respectively. Every weight sits in VRAM, pinned RAM, or unified memory, so nothing streams from disk and nothing swaps.
- Five everyday devices (two PCs, a MacBook, two iPhones) were linked to run a large AI model together.
- The combined system generates text at 37-39 tokens per second—fast enough for real-time chat.
- This approach could let people run powerful AI on existing hardware, saving money and keeping data local.
Why It Matters
This could make advanced AI more affordable and private by using devices you already own instead of expensive servers.