128GB RAM plus dual 3090s: is upgrading a home AI server still worth it?
LankyGuitar6528 · reddit · 2026-11-05
A Reddit user debates whether to upgrade an existing home AI server (32GB RAM + one 3090) by spending $800 on 4x32GB for 128GB memory and adding a cheap second 3090 on a riser, or buying a new machine outright, given recent hardware trends like N1X chips and unified memory. The thread asks the community for a verdict on the best bang-for-buck path to running the largest possible local model.
More from Infra
- Anthropic's IPO filing lists compute scarcity as key risk; WSJ says tech giants win — pstAsiatech · 2026-11-05
- Let there be light: novel AI framework runs neural networks on optical hardware — striketheviol · 2026-11-05
- OpenAI pledges to cover full electricity and grid costs of future data centers — VraserX · 2026-11-05
- AMD's AI stack comes of age: dev runs full RL LoRA training loop on a Framework desktop — qtnx_ · 2026-11-05
- Hugging Face to Host 'Local AI 301' Livestream With 4 Talks on Quantization, llama.cpp and vLLM — huggingface · 2026-11-05
- Data centers are hated for fixable reasons — noise, looks, water, energy — yet no one tries — dansitu · 2026-11-05