RAMDeck Promises Bigger Local AI Models by Sharing Work Across Devices

The Short Version

TradeMAV is seeking crowdfunding for RAMDeck, a hub intended to help several household devices run an AI model together. TechRadar reports a $109 early-backer reward and says the underlying engine uses llama.cpp’s freely available RPC technology. The proposed benefit is fitting a model that exceeds one device’s memory.

Why It Matters

The devices keep their own memory. Different machines hold different model layers and pass intermediate results over the network while generating an answer, according to the report. This is distributed execution, rather than a physical merger of RAM into one faster memory system.

TradeMAV’s pitch adds coordination around that process, including finding devices automatically, assigning layers and providing a dashboard. For a user, the question is whether that convenience justifies a separate hub when the underlying software can already be used without buying it.

The Bigger Picture

The company’s published results show why capacity and responsiveness need separate attention. TechRadar cites a 27-billion-parameter dense model producing 1.92 tokens per second across four devices. Faster headline demonstrations used different models and hardware, including an RTX 3060, so they are not interchangeable performance promises.

These are vendor figures described in the report, rather than independent measurements of a delivered RAMDeck product. A model successfully fitting into memory does not, by itself, establish that its response speed will suit your work.

What to Watch

The hub’s processor, memory, ports and power draw remain undisclosed in the account. Those specifications and tests of shipping hardware would make the convenience claim easier to judge. The $109 figure is a crowdfunding reward, so delivery and final performance should still be treated as promises.

Would you accept slower AI responses to run a larger model on devices you already own?