[{"id":"IL-0023","title":{"en":"A training run crashed with an NCCL timeout after one InfiniBand port went down once and recovered","ko":"InfiniBand 포트 하나가 한 번 내려갔다 복구되면서 7시간째 돌던 학습 작업이 NCCL 타임아웃으로 죽었다"},"org":"Databricks","org_slug":"databricks","date":"2026-07-01","severity":"unknown","root_cause":"dependency-failure","url":"https://benchlet.com/incidentlib/app/cases/IL-0023","shared_components":["infiniband","nccl","pytorch"]},{"id":"IL-0126","title":{"en":"The OPT-175B training run restarted more than 40 times in two weeks as cloud GPU nodes failed with ECC errors, replacement nodes came back faulty, and checkpoint downloads hung","ko":"OPT-175B 학습이 클라우드 노드의 ECC 메모리 오류, 같은 결함을 안고 돌아온 교체 노드, 멈추는 체크포인트 다운로드 때문에 2주 동안 40번 넘게 재시작했다."},"org":"Meta AI","org_slug":"meta-ai","date":"2021-12-03","severity":"unknown","root_cause":"dependency-failure","url":"https://benchlet.com/incidentlib/app/cases/IL-0126","shared_components":["infiniband","nccl"]},{"id":"IL-0133","title":{"en":"A storm shut down cooling at Nebius's us-central1 data center, and a disk hot-plug bug kept Managed Kubernetes control planes down for hours after power returned","ko":"폭풍으로 데이터센터 냉각이 멈춰 us-central1 리전 전체가 내려갔고, 디스크 핫플러그 버그로 Managed Kubernetes 컨트롤 플레인 약 140개의 복구가 길어졌다."},"org":"Nebius","org_slug":"nebius","date":"2026-08-19","severity":"unknown","root_cause":"dependency-failure","url":"https://benchlet.com/incidentlib/app/cases/IL-0133","shared_components":["infiniband"]},{"id":"IL-0134","title":{"en":"A power event at Lambda's AUS01 data center rebooted 246 GPU nodes, all spine switches and the VAST storage, failing customer workloads and leaving stale NFS mounts on VMs","ko":"AUS01 데이터센터에 전력 문제가 생겨 GPU 노드 246대와 스파인 스위치, VAST 스토리지가 한꺼번에 재부팅됐고, 워크로드가 실패하고 NFS 마운트가 끊겼다."},"org":"Lambda","org_slug":"lambda","date":"2026-02-14","severity":"sev1","root_cause":"dependency-failure","url":"https://benchlet.com/incidentlib/app/cases/IL-0134","shared_components":["infiniband"]},{"id":"IL-0021","title":{"en":"GKE GPU nodes on Container-Optimized OS could not download NVIDIA drivers (403 Forbidden) and came up without working GPUs","ko":"GKE 의 COS 노드가 NVIDIA GPU 드라이버를 403 Forbidden 으로 받지 못해 새로 만든 GPU 노드를 쓸 수 없었다"},"org":"Google Cloud","org_slug":"google-cloud","date":"2024-03-12","severity":"unknown","root_cause":"dependency-failure","url":"https://benchlet.com/incidentlib/app/cases/IL-0021","shared_components":["gpu-driver"]}]