LocalLLM Rigs

Memory bandwidth explained

Memory bandwidth is the rate at which a processor can move data to and from memory, measured in GB/s. For local LLMs it is often the single most important speed spec. Each generated token requires reading the model's weights, so tok/s scales roughly with bandwidth. Our catalog spans 89.6 GB/s to 614 GB/s. The MINISFORUM UM870 Slim, UM890 Pro, AI X1 Pro-370 and the GEEKOM A9 Max all use dual-channel DDR5 at 89.6 GB/s. The AMD Ryzen AI Halo reaches 256 GB/s. Apple's unified-memory machines range from 170 GB/s (Mac mini M6) through 307 GB/s (Mac mini M5 Pro) to 614 GB/s (Mac Studio M5 Max). Bandwidth is fixed by the platform. Adding more RAM raises capacity, not bandwidth.