Researchers rented 51,500 GPUs across AWS, Azure and GCP for a two-hour cloudburst experiment

mcraddock · x · 2026-09-03

The San Diego Supercomputer Center and Wisconsin's WIPAC ran a GPU cloudburst experiment in 2019, marshalling essentially all for-sale GPUs across AWS, Azure and Google Cloud to peak at 51,500 GPUs over roughly two hours, funded by an NSF EAGER grant.

The compute powered simulations for the IceCube Neutrino Observatory at the South Pole. A chart of the 200-minute burst shows how fast cloud infrastructure could scale for short, intense scientific workloads. The poster wistfully calls it 'the good old days of cloud GPU access.'

Original post →

More from Infra

Infra channel →