DIY AI server packs Blackwell, EPYC

This is less a “home server” than a compact inference appliance disguised as one.

• –The 96GB GPU is the headline feature: enough VRAM to make local LLM work far less compromise-heavy.
• –EPYC plus 512GB ECC RAM suggests this machine is built for sustained throughput, not just benchmark flexing.
• –The Supermicro board and Ubuntu combo signal a practical, server-first setup with good odds of stable driver and container support.
• –For local AI developers, this is the sweet spot where one machine can host models, vector stores, and supporting services without constant resource juggling.
• –The main constraint is cost, not capability; this is premium hardware aimed at people who value time saved more than dollars spent.