Lawyer Constructs Private AI Server with Eight Used GPUs

• A lawyer built a private AI server using eight used 2018-era Tesla V100 GPUs.
• The server includes two 4-card SXM2 NVLink boards, four PEX switches, and SlimSAS cables.
• The server runs the open DeepSeek V4.1 Flash model, processing up to 215 tokens per second.
• The lawyer built the server to ensure case files never leave the office.
• The project encountered an issue with a burnt power connector.