Local LLM inference/training workstation. RTX Pro 6000 Blackwell 96GB VRAM. Priced at current market for the GPU alone plus the supporting build.

If you know why 96GB of VRAM on a single card matters for 70B+ parameter models at FP8, this is for you. If you're comparing this to a gaming PC, it isn't one.

• NVIDIA RTX Pro 6000 Blackwell Max-Q Edition, 96GB GDDR7 with ECC, PCIe 5.0. Runs Llama 70B class models fully in VRAM, no quantization compromises, no multi-GPU tensor parallelism overhead\
• AMD Ryzen 9 9950X (16C/32T)\
• 128GB DDR5 (Kingston FURY Beast KF552C40BWK4-128), enough headroom for CPU offload and large context KV cache spillover\
• 2x Samsung 9100 PRO 1TB PCIe 5.0 NVMe, fast model loading and checkpoint writes\
• ASUS ROG Crosshair X870E Hero, AM5\
• Seasonic VERTEX PX-1200, ATX 3.1 / PCIe 5.1 native 12V-2x6\
• Arctic Liquid Freezer III Pro AIO\
• Fractal North XL full tower

Brand new, moved my workloads to cloud GPUs. Serious inquiries only. If your first question is "will it run games," save us both the time.

The GPU alone now costs more than most people’s cars. Look up current RTX Pro 6000 Blackwell pricing before reacting.