€550 Used HP Z640 With Dual RTX 3060s: Reader Asks for MoE Local LLM Build Sanity Check
ElekDn · reddit · 2026-10-20
A Reddit user plans to buy a used HP Z640 for €550 (2x E5-2680 v4, 64GB ECC 2400, bundled Quadro M6000 24GB, 6-hour clean MemTest86) and add two RTX 3060 12GB cards to run Qwen MoE models, llama.cpp, PyTorch CV training, PyBullet RL, LLM LoRA fine-tuning, and MiniMax video generation.
Open questions for the community: whether dual-socket causes issues for llama.cpp and MoE offload, if the M6000 is worth keeping as a third CUDA 12 card, whether 86°C CPU temps are normal, and if adapters plus power-limiting the 3060s is the right approach given the PSU's two 6-pin cables and HP's 150W-per-card limit with two CPUs.
More from Infra
- The Local Computing Rebellion: Subscription Fatigue and Forced AI Drive a Shift Back On-Premises — DavidLinthicum · 2026-10-20
- DeepSeek also open-sources deepseek-recipe, replicating official API token encoding logic — deepseek_ai · 2026-10-20
- DeepSeek open-sources DeepInfer, the production inference framework behind DeepSeek-V4.1-Flash — deepseek_ai · 2026-10-20
- Scott Hanselman serves local inference to his whole house from a Surface Laptop Ultra — lee_stott · 2026-10-20
- DeepSeek-V4.1-Flash quantized to 2-4 bit mixed precision for dual RTX PRO 6000 with 0.1306 KL divergence — TheZachMueller · 2026-10-20
- $4K AMD Strix Halo Mini PC with 128GB Unified Memory Runs Full-Repo Local LLM Workflows — MooseEfficient2151 · 2026-10-20