• Update to DCGM 4.5.3 and DCGM Exporter 4.8.2.
• Improve GPU health metrics, including reporting GPU-wide health incidents such as fallen-off-bus XIDs.
• Make /debug/pprof profiling endpoints opt-in via --enable-pprof / DCGM_EXPORTER_ENABLE_PPROF.
• Add PodMapper informer caching for Kubernetes pod mapping (#626) (@jaeeyoungkim).
• Add per-process GPU metrics for time-sharing and MIG (#594) (@krystiancastai).
• Make Helm priorityClassName configurable with explicit defaults (#444) (@runzhliu).
• Add MIG device support for HPC job labels (#602) (@jay-mckay).
• Update go-dcgm field metadata handling, deprecated field alias resolution, health constants, policy registration handling, and version info APIs.
• Document IPv6 address formats for remote hostengine and metrics listen addresses.
• Refresh dependencies, container base images, Docker image references, Helm chart values, Kubernetes manifests, and tests for this release.

Contributors

• @runzhliu
• @jay-mckay
• @jaeeyoungkim
• @krystiancastai

• Update to DCGM 4.5.2, latest Go 1.24, and base containers
• Fix distroless symlink issue
• Fix for parsing blank XIDs
• Fix for nvlink entities starting at offset 1

• Update to DCGM 4.5.1
• Enabled monitoring of GPU bind/unbind events and automatic reloading (@nvvfedorov) - beta
• Sync default metric watchlist for docker and helm (@faizan-exe)
• Fix health endpoint behavior (@Alja9)
• Increase default memory limit to 512Mi (@faizan-exe)
• Make scrapeTimeout configurable (@faizan-exe)
• Fix P2P Status mappings (@wkd-woo)

NOTICE: Helm chart now uses distroless container by default

Contributors

• @Alja9
• @wkd-woo
• @faizan-exe
• @nvvfedorov

• Update Go-DCGM
• Update XID error texts based off "XID errors v580" (#588)
• FIX: fix for vailation time us->ns (#589)
• feat: updating readme to include OpenObserve blog and dashboards for ... (#580)

• Security improvements in startup and pipeline @nvvfedorov
• Crash if collector cannot be initialized under startup validation (#578) @daveoy
• Track unallocated GPUs in DRA (#570) @JiangJiaWei1103
• Fix unbounded label cache size (#574) @andrew-leung

Contributors

• @andrew-leung
• @daveoy
• @JiangJiaWei1103
• @nvvfedorov

• Add allow list for pod label filtering (#564)
• handle uninitialized map (#563)
• Add hostPID field to values.yaml and DaemonSet template for Helm (#503)
• feat(dcgm-exporter): add option to fail on nvml provider init error (#557)
• Improved support for GPU NvLink monitoring

• Follow FHS convention for logging: #556
• Fix: enable metrics without pods using kubernetes-enable-{dra,virtual-gpus} #554
• Add disable startup validate flag #555
• hpc: reduce error logging if jobs directory does not exist

• Update to DCGM 4.4 and Cuda 13.0
• Kubernetes UID support (@andrew-leung)
• Create distroless container target

Contributors

• @andrew-leung

• Update To DCGM 4.3.1
• Update podapi for DRA
• Enable DCGM_EXP_P2P_STATUS for reporting GPU peer-to-peer nvlink status
• Fix for empty HPC directory
• Enable InitContainer support

DCGM-Exporter 4.2.3-4.2.0

• [ISSUE-512] Added a new debugging facility to dump runtime objects into files
• Kubernetes pod label support