Follow us
Add us as a preferred source on Google
- TradeMAV’s RAMDeck Hub is crowdfunding on Indiegogo, listed at $109 for early backers
- The engine underneath is llama.cpp’s RPC backend, which anyone can already run for free
- Its fastest demos rely on an RTX 3060 and mix of expert models, while denser models manage around two tokens per second
With DDR5 now costing about five times what it did in mid-2025, a small company called TradeMAV believes the answer to running larger AI models at home is not to buy more RAM, but to borrow the memory already sitting in your other devices.
The argument, which is difficult to rule against, is set up in the form of RAMDeck, which is TradeMAV crowdfunding on Indiegogo, a small low-power hub that connects Windows PCs, Macs, Linux machines and Android phones on a home network in a memory pool, so that a model too big for each individual device can be divided over all.
The campaign uses fixed funding against a $10,000 goal and closes on November 8, 2026, according to the company’s launch video, while a cost comparison table in TradeMAV’s own RAM Pooling Guide lists the Hub at $109 as an early bird reward.
Latest Videos FromTechRadar
An old trick in a new box?
The timing is interesting, to say the least: German retail data from 3DCenter shows DDR5 at 486% of its July 2025 price by mid-August; Samsung told analysts that the shortage will persist until 2028; and the chairman of the SK Group warned it could run until 2030.
All this creates a perfect storm, and TradeMAV recognizes it; its origin story leans on the same squeeze, claiming that a 64GB DDR5 kit went from around $191 in August 2025 to $1,118 a year later. That is broadly in line with our own findings that average 2x32GB DDR5-5600 kit prices will roughly quadruple to around $800 in the last quarter of 2025 alone.
Look up
What RAMDeck actually does is less exotic than “Pooling RAM” suggests. Nothing is merged at the hardware level. Instead, the model’s layers are divided among machines, each holding its part in local memory, and only intermediate results pass over the network as each token is generated.
TradeMAV’s public GitHub repository is honest about the plumbing involved: The node agent starts llama.cpp’s ggml RPC server on each contributing device; the README says that any compatible coordinator can run it via the standard llama.cpp RPC protocol, and the suggested way to confirm that it works is a stock llama-cli command.
TradeMAV’s contribution is the orchestration wrapped around it: automatic discovery, layer placement that fills fast GPUs before slower system RAM, error recovery, a dashboard, and an OpenAI-compatible API, according to its features page.
While the process is detailed, the other components of the hub detail are missing; None of TradeMAV’s published pages list its processor, memory, ports, or power draw. His own pooling guide admits that if you’re comfortable with terminal scripts, “the free open-source DIY route is ideal.”
Pooling devices for inference is not new. Exo Labs’ open source Exo has offered automatic device discovery in a drop-in API via computers, phones, and single-board computers since at least early 2025. At the premium end, Apple’s new M5 Ultra Mac Studio supports clustering via Thunderbolt 5 with RDMA, and Apple claims four machines deliver up to three times the inference speed of one.
RAMDeck’s distinctive pitch is broad – its FAQ promises a Windows CUDA desktop, an Apple Silicon Mac, an old Intel Mac, a Linux box and an Android phone in the same cluster, something that makes llama.cpp’s mix of backends possible.
TradeMAV is unusually frank about the performance trade-off versus something like a Mac Studio configuration. His story page warns that “pooled memory trades speed for capacity,” and his benchmark write-up tells readers chasing raw performance to buy a higher VRAM GPU instead.
The headline results are on a Beelink Mini PC with an RTX 3060 and on mix of expert models that activate only about 3 billion parameters per token: 39 tokens per second for Qwen3 coder 30B over the Mini PC and an M4 Mac Mini, and 16 tokens per second for a 40B REAP variant of coder QNwen3 coder. A 7B model that hits 43 tokens per second was, by TradeMAV’s own description, running on the 3060.
Density models tell a different story: a 27B dense model managed 1.92 tokens per second across four devices, and OpenAI’s gpt-oss-120b ran at 1.37 tokens per second on a six-device cluster. A 120B model on another five-machine setup achieved 3.1 tokens per second, after taking about seven minutes to initialize. That beats a crash outside of memory, which is TradeMAV’s pitch, but it’s extremely slow.
In a nutshell, RAMDeck’s engine wraps free software, the specifications of the hub are not known, and some published figures indicate limitations when applying this approach. For tinkerers, llama.cpp’s RPC mode in Exo costs nothing to try first. For everyone else, $109 buys convenience and, as with all crowdfunded hardware, a promise rather than a product.
Disclaimer: We do not recommend or endorse any crowdfunding project. All crowdfunding campaigns carry inherent risks, including the possibility of delays, changes or non-delivery of products. Potential backers should carefully evaluate the details and proceed at their own discretion.
Follow TechRadar on Google News in the add us as a preferred source to get our expert news, reviews and opinions in your feeds.