NH
A great all-round local AI computer
Exactly what I needed to run local AI. Installed TailScale on the Spark and my local laptops, and now I can SSH and access the Spark from anywhere using the TailScale network without needing to expose my Spark with a public IP... so strongly suggest others do this also and NVIDIA has a playbook on how to install and configure TailScale. Do use the NVIDIA playbooks as they are very helpful and provide lots of good ideas. Also setup my DGX without connecting any mouse, keyboard, or monitor... just follow the setup instructions from the NVIDA DGX site. It does get very hot if you're running a big model for a long task, but I've not had any sudden reboots because of the heat. The fans seem to do an adequate job of dissipitating heat quite efficiently. Finally I used Claude Code to help me configure the local AI models using Ollama or llama.cpp and then accessing via OpenCode. Claude does a great job of tuning the models for optimum memory load and that has cut down the load time for some of these models significantly. Also it's good with reading over the model sheet in Hugging Face and then suggesting which would be the best model to install for your desired application.