📋 How to run eight #NVIDIA #DGXSpark boxes as a headless #LLM inference cluster: two groups of four, each serving #DeepSeek V4.1 Flash split four ways over RoCE. Stock kernel and driver, all changes are settings, not patches.⚡ A GB10 has no power-limit setting, the GPU clock is the only lever. At stock clocks the rack averaged 1,717 W and tripped its 15 A strip after ~8 minutes. A 2,200 MHz cap brought it to 1,345 W (−22%) with no drop in decode speed.🧵👇
📋 How to run eight #NVIDIA #DGXSpark boxes as a headless #LLM inference cluster: two groups of four, each serving #DeepS...
