A low-carbon computing platform from your retired phones

Details

The carbon footprint of computing splits into two parts: operational emissions from energy use, and embodied carbon from manufacturing. Operational carbon gets the attention, but manufacturing new servers is a significant and harder-to-fix source of emissions. Researchers at UC San Diego, with Google's support, are tackling embodied carbon directly. Their approach — 'phone cluster computing' — extracts the motherboards from retired consumer smartphones, assembles them into clusters, and redeploys them as a general-purpose cloud computing platform. The plan is to build a datacenter from 2,000 retired Pixel phones, offering low-cost cloud computing to hundreds of researchers and students. The performance case is credible. Per SPEC benchmarks, a 2023 Pixel Fold's performance cores match or beat the per-core throughput of a modern ASUS RS720A-E11 server. The trade-off is density: phones have 8-12GB RAM and a handful of heterogeneous cores versus dozens of server cores with far more memory. The research focuses on identifying workloads that fit within these constraints. The practical question — what kinds of jobs can actually run on phone clusters — remains partly open. But for academic computing, latency-insensitive batch jobs, and certain ML inference tasks, the approach could be both technically viable and meaningfully greener than buying new hardware.