Together launches GPU Auto Repair to fix failing AI nodes faster

Together has made GPU Auto Repair generally available, detecting broken nodes and fixing them without unnecessary reboots. The system monitors both customer VMs and the host, responding to Xid faults, double-bit ECC errors, PCIe failures and unstable InfiniBand links.

• The system picks the lightest repair for each fault and escalates only if it fails
• Customers can set repairs to run automatically or wait for approval
• Preview support for external detection systems via node-label contracts is coming