Is upgrading from 2x to 4x RTX 3090 worth it for local LLM work?
fgoricha · reddit · 2026-09-22
A local LLM hobbyist shares his rig (X299, i9-7940X, 2×3090 FE, 1600W PSU), running speech-to-text, Qwen models for image labeling to build YOLO datasets. He's weighing adding a third or fourth 3090 — PCIe lanes allow it, but a fourth card needs another PSU — and asks what 3-4 cards would unlock over 2. Not interested in cloud APIs.
More from Infra
- Alibaba roadmap: Qwen explores RSI, new models up to 10T params, 20GW datacenters by 2032 — teortaxesTex · 2026-09-22
- Alibaba Plans AI Model With 5-10 Trillion Parameters, Unveils New Chip — tengo_harambe · 2026-09-22
- OpenAI removes Ultrafast tier from GPT-5.6 Sol in Codex, fueling GPT-6 Sol rumors — imjustnewatai · 2026-09-22
- Meituan's unified recommendation foundation model MTFM lifts orders up to 6.68%, cut inference cost 24% — 美团技术团队 · 2026-09-22
- Qwen-Image local on a 24GB MacBook Pro takes 5-6 minutes per image — vista8 · 2026-09-22
- Cerebras CEO on Jensen Huang: a decade trading as 'nobody' before Nvidia made it — rohanpaul_ai · 2026-09-22