LLM Inference Workstation — 4x RTX6000 Blackwell Pro Max-Q (384GB VRAM) / Threadripper Pro 7985WX / WRX90E-SAGE SE / 512GB RAM / 1700W

A forum user on Level1Techs detailed a new LLM inference workstation built with four PNY NVIDIA RTX PRO 6000 Blackwell GPUs (384GB VRAM), an AMD Ryzen Threadripper Pro 9975WX CPU, an ASUS Pro WS WRX90E-SAGE SE motherboard, 512GB DDR5-6400 ECC RAM, and a 1700W FSP Cannon Pro power supply, purchased with an Nvidia inception discount before recent price hikes. The user cited IP sovereignty, privacy, censorship, and avoiding corporate cloud models as reasons for the investment, and sought community guidance on configuration and agentic-coding setups.

Hey guys - New here, but came across this thread so thought it might be good to jump in. Nice spec machine @ambientlight https://forum.level1techs.com/u/ambientlight . I’m hoping you guys can point me to any discord or groups with people with similar spec machines that can share config, test results, best setups for agentic-coding, etc. I’m finding that I’m trying to balance time building with staying updated with what’s new and then end up in perpetual rabbit holes trying to optimise my configuration. Life was way easier just using a cloud model and skipping all of this, but I could not accept the compromise, i.e. IP sovereignty, privacy, censorship, feeding the corporates, etc so I decided to invest in some hardware luckily I managed to get Nvidia inception discount and paid just before the two recent price hikes . Anyway, my spec is below. Feel free to abuse me if I could have done something better: 1x AMD Ryzen Threadripper Pro 9975WX CPU 1x ASUS Pro WS WRX90E-SAGE SE Motherboard 8x 64GB DDR5-6400 ECC R-DIMM 4x PNY NVIDIA RTX PRO 6000 Blackwell GPUs 1x FSP Cannon Pro 2500W Power Supply 1x 360mm AIO Liquid Cooler 3x Phanteks M25 140mm Fans 1x Antec 900 Full Tower Case 1x Samsung 9100 Pro 4TB M.2 SSD 1x Kingston DC300ME 7.68TB U.2 SSD