Enthusiast Maxes Out Local LLM Rig with RTX 5090 and RTX PRO 5000, Reaching 80GB VRAM
English summary
A LocalLLaMA community member completed a multi-GPU build using an existing RTX 5090 and a newly acquired RTX PRO 5000, achieving 80GB of total VRAM. The 9950X3D system also includes 192GB RAM and 17TB storage, powered by a 1300W PSU. The user originally planned to buy an RTX PRO 6000 for $8.5K with a hoped-for NVIDIA Inception discount, but after a 3-month wait the application was rejected and the product price surged to $13.5K. They instead purchased the last available RTX PRO 5000 in their country with the saved funds. The rig is now used for large quantized LLMs (Q8) and multi-GPU ComfyUI workflows.
Chinese summary
一位 LocalLLaMA 社区成员用已有的 RTX 5090 和新购入的 RTX PRO 5000 搭建了双卡主机，总显存达 80GB。该主机搭载 9950X3D 处理器、192GB 内存和 17TB 存储，由 1300W 电源供电。用户原本计划以 8500 美元购买 RTX PRO 6000 并申请 NVIDIA Inception 计划折扣，但等待 3 个月后申请被拒，此时产品价格已涨至 13500 美元。最终他们用准备的钱在本国买到了最后一块 RTX PRO 5000。现在该主机用于运行大型 Q8 量化 LLM 和多 GPU ComfyUI 工作流。
Key points
Dual-GPU system combines RTX 5090 and RTX PRO 5000 for 80GB total VRAM.
双 GPU 系统由 RTX 5090 和 RTX PRO 5000 组成，总显存 80GB。
Original RTX PRO 6000 purchase plan at ~$8.5K was abandoned after NVIDIA Inception rejection and a price increase to $13.5K.
原本购买 RTX PRO 6000（约 8500 美元）的计划因 NVIDIA Inception 申请被拒及价格涨至 13500 美元而放弃。
Acquired the last available RTX PRO 5000 in their country using saved funds.
用准备的钱在本国买到了最后一块 RTX PRO 5000。
System specs: 9950X3D CPU, 192GB RAM, 17TB disk, 1300W PSU; used for Q8 quantized LLMs and multi-GPU ComfyUI.
主机配置：9950X3D CPU、192GB 内存、17TB 硬盘、1300W 电源；用于 Q8 量化 LLM 和多 GPU ComfyUI。