Microsoft Patents System to Train AI on Idle Consumer GPUs
Microsoft has filed a patent application outlining a framework to divide AI model training into smaller tasks and distribute them to opted-in users' idle consumer GPUs in exchange for credits. This approach could offer a crowdsourced alternative to expensive centralized data centers, potentially lowering the massive infrastructure costs associated with training modern AI models. The application details an application-based opt-in system where rewards are paid in credits, though specific payout rates and solutions for home internet latency are not yet specified.
## BACKGROUND
Volunteer computing is a form of distributed computing where individuals donate their computer's unused processing resources to research projects, often in exchange for points or credits. Modern desktop computers and consumer GPUs frequently sit mostly idle during routine tasks, leaving massive amounts of untapped computational power available.