Denpex is a purpose-built cluster diagnostics and reliability platform for high-performance computing and AI infrastructure. Designed specifically for H100 and B200 GPU clusters, Denpex isolates hardware failures before they cascade through your training runs. When a node fails, traditional monitoring tools often lose visibility or flood your logs with secondary network errors. Denpex takes a different approach: it utilizes offline diagnostics to pinpoint the exact failing component without relying on the primary cluster network.
A core capability of Denpex is its comprehensive Xid error analysis. It automatically decodes and correlates NVIDIA Xid errors across thousands of GPUs, distinguishing between transient software faults and permanent hardware degradation. Furthermore, Denpex provides strict NCCL timeout isolation. Instead of a single GPU stall hanging an entire distributed training job and wasting thousands of dollars in compute time, Denpex identifies the unresponsive rank, isolates the fault, and provides actionable recovery steps immediately.
For machine learning engineers and system administrators managing massive AI workloads, Denpex reduces mean time to resolution from hours to minutes. It ensures your infrastructure spends more time training and less time waiting for manual triage.
