AIXDA1 day ago
Old GPUs outperform a $2000 card for AI work
1 min read
An XDA writer runs local AI models on salvaged RTX 3080 Ti GPUs using Mixture-of-Experts architecture. MoE models use a router and experts design that reduces VRAM needs, allowing the 12GB card to run Qwen3.6. The writer sees no reason to upgrade to a new $2000 card.
Level
Hype check
Tap to vote and see what everyone thinks.
