tech
This crowdfunded $109 hub wants to pool the RAM in your old PCs, Macs, and phones to run bigger AI models — but the engine underneath is software you can already get for free
A new crowdfunded device called RAMDeck promises to let home users run larger artificial intelligence models by linking the unused memory inside Windows PCs, Macs, Linux machines and Android phones over a home network.
Latest updates
This crowdfunded $109 hub wants to pool the RAM in your old PCs, Macs, and phones to run bigger AI models — but the engine underneath is software you can already get for free
A new crowdfunded device called RAMDeck promises to let home users run larger artificial intelligence models by linking the unused memory inside Windows PCs, Macs, Linux machines and Android phones over a home network. TradeMAV is offering the low-power hub on Indiegogo for 109 dollars as an early-bird reward, with the campaign closing on November 8, 2026. The pitch arrives as DDR5 memory prices have surged, with German retail data showing costs at nearly five times their July 2025 level by mid-August. The system does not physically merge memory; instead, it splits a model’s layers across devices and sends only intermediate results over the network. TradeMAV’s own benchmarks show dense models running as slowly as about two tokens per second, while faster results rely on a dedicated RTX 3060 graphics card. The underlying engine is the free llama.cpp RPC backend, meaning technically comfortable users can already try the same approach at no cost. For everyone else, the 109 dollars buys convenience and orchestration, but as with any crowdfunded hardware, backers are funding a promise rather than a finished product.
This crowdfunded $109 hub wants to pool the RAM in your old PCs, Macs, and phones to run bigger AI models — but the engine underneath is software you can already get for free
A new crowdfunded device called RAMDeck promises to link Windows PCs, Macs, Linux machines, and Android phones on a home network into one shared memory pool, letting a model too large for any single device run across all of them. The Indiegogo campaign lists the hub at $109 for early backers, with a fixed funding goal of $10,000 and a closing date of November 8, 2026. The pitch arrives as DDR5 memory prices have climbed to roughly five times their mid-2025 levels, with industry warnings that the shortage could persist until 2028 or beyond. TradeMAV is candid that the underlying engine is llama.cpp’s free RPC backend, and its own guide says do-it-yourself users comfortable with terminal scripts can skip the hardware entirely. Performance varies sharply: a mixture-of-experts model reached 39 tokens per second with a fast graphics card, while a dense 120-billion-parameter model managed just over one token per second on six devices. For an ordinary home user, the real-world trade-off is clear: you can avoid buying expensive new RAM, but you may wait minutes for a single response. Backers are funding a promise rather than a finished product, and the company has not yet published the hub’s processor, memory, ports, or power draw.