Used 64-core Threadripper PRO 3995WX
retained text
Found a used 64-core Threadripper PRO 3995WX workstation locally — worth buying for a multi-GPU LLM server? I found this workstation listed on my local secondhand marketplace in Iraq. Specs \- CPU: AMD Ryzen Threadripper PRO 3995WX (64 cores / 128 threads) \- Motherboard: ASUS Pro series (exact model unknown) \- RAM: 224 GB DDR4-3200 RGB \- Storage: 128 GB SSD \- PSU: ASUS ROG Strix Platinum 850W \- GPU: GTX 1050 Ti The seller hasn't provided further details or a price yet. My intended use: I'm planning a local LLM inference server, potentially with 4–8 RTX 3090 GPUs, using engines such as llama.cpp, vLLM, or SGLang. I currently have an RTX 4090, RTX 3090 Ti, and 128 GB DDR4 RAM. I'm particularly interested in running large MoE models, multi-GPU inference, and possibly CPU/RAM offloading. Questions: 1. What would be a fair used price for this complete workstation in USD? 2. Could the motherboard be the ASUS Pro WS WRX80E-SAGE SE WIFI, which has seven PCIe 4.0 x16 slots? 3. Would my existing consumer DDR4-3200 UDIMM RAM be compatible, or does this platform require ECC RDIMM? 4. How practical is running 4–7 GPUs on this motherboard using risers and external power supplies? 5. Is the 3995WX still a good choice for large MoE inference and CPU offloading, or would a cheaper EPYC platform make more sense? 6. Does the 224 GB RAM configuration raise any red flags? I'm mainly interested in PCIe lanes, memory bandwidth, upgradeability, and cost-effectiveness rather than raw CPU compute. Would you buy this platform in 2026, and what's the maximum you'd pay for it?
document fetched 2026-11-01 00:38 · canonical https://reddit.com/r/LocalLLaMA/comments/1xkdpv3/used_64core_threadripper_pro_3995wx