ASCI Q supercomputer crashed every 6.5 hours until a neutron beam exposed cosmic rays as the culprit

lauriewired · x · 2026-09-29

lauriewired revisits the reliability crisis of ASCI Q, the world's #2 supercomputer in 2002-2003: it failed every 6.5 hours on average, full-machine reboots took up to 8 hours, and utilization sat at only 60%, with rare "few good weeks" above 85%.

Engineers suspected cosmic rays causing uncorrectable bit flips. At Los Alamos they pointed their own neutron beam at the machine—5 seconds of beam time equaled 6+ years of normal sun exposure—and traced the failures to missing correctable parity in the L2 cache of the Alpha CPUs. Later machines adopted ECC-protected L2 plus better software fallbacks and checkpointing.

Takeaway: when pushing the frontier, you sometimes have to work with very unreliable systems.

Related event: ASCI Q Supercomputer Crashed Every 6.5 Hours(2 posts)→

Original post →

More from Fun

Fun channel →