Hey everyone,

Last time I built a PC was around 2010, so I'm basically a boomer in PC building years. Need your eyes on this before I pull the trigger.

Use case: Running large language models locally (14B+ parameter models with RAG setup). Privacy is important so everything needs to stay on-premise. Will be processing a lot of documents and need the VRAM for inference.

Budget: Not really a concern, just want it to work reliably.

The Build:

PCPartPicker Part List

GPU: NVIDIA RTX 5090 32GB (if I can find one at reasonable price) OR RTX 4080 Super 16GB as backup

CPU: AMD Ryzen 9 7950X

Motherboard: ASUS ROG Strix X670E-E Gaming WiFi

RAM: 128GB DDR5-6000 (4x32GB) G.Skill Trident Z5

Storage:

• Samsung 990 Pro 2TB NVMe (OS + models)
• Samsung 870 EVO 4TB SSD (data)

PSU: Corsair RM1000x 1000W 80+ Gold

CPU Cooler: Noctua NH-D15

Case: Fractal Design Torrent

Questions:

1. Is 1000W enough for the 5090 or should I go 1200W? (Spec says 575W TDP + 1000W recommended PSU)
2. Any compatibility issues I'm missing?
3. Overkill on the RAM or justified for large vector databases?

Thanks!