作者位次 #6/19OSDI 2026Oral· Operating Systems & Distributed Systems
SDCs in the Wild: Characterizing and Diagnosing SDC-Defective GPUs in Production LLM Training (Operational Systems)
Wenxin Zheng · Wenxiao Wang · Yun Zhang · Mingcong Han · Bin Xu · Jinyu Gu · Xingda Wei · Haibo Chen · Zuquan Song · Gaohong Liu · Yucheng Nie · Zhe Nan · Zhuolin Zheng · Huan Yu · Shuguang Wang · Ziming Zhou · Hang Zhu · Wencong Xiao · Xin Liu
摘要
Silent Data Corruption (SDC) has emerged as a critical reliability bottleneck in Large Language Model (LLM) training, where hardware faults are frequently indistinguishable from software anomalies. While standard industry practice relies on synthetic microbenchmarks for fault isolation, our exper...