DCGM Exporter Release Notes#

This page reproduces the changelists published with the official DCGM Exporter GitHub releases. Each entry links to its original release page.

4.8.4#

Released:

2026-09-18

Tested with DCGM:

4.6.1

Source:

4.8.4 on GitHub

Changelist#

Runtime and performance#

  • Release DCGM Exporter 4.8.4 with DCGM 4.6.1 and go-dcgm v1.4611.2.

  • Add configurable DCGM field-watch retention limits through CLI, environment, and YAML configuration.

  • Add the version 2 YAML schema with plural collections.

  • Add configurable limits and request coalescing for concurrent /metrics scrapes.

  • Reduce allocations during metric gathering, transformation, and rendering.

  • Support metric configurations containing more than 128 DCGM fields by batching reads and splitting oversized field groups.

  • Add optional GPU bind/unbind detection that resets DCGM and NVML, then rebuilds metrics for the new GPU topology.

  • Honor LD_LIBRARY_PATH when locating DCGM libraries.

Metrics and Kubernetes#

  • Add opt-in Go runtime, process, and HTTP handler metrics through --enable-exporter-metrics, DCGM_EXPORTER_ENABLE_EXPORTER_METRICS, or YAML configuration.

  • Add opt-in GPU fabric health fields: DCGM_FI_DEV_FABRIC_HEALTH_SUMMARY and DCGM_FI_DEV_FABRIC_HEALTH_MASK.

  • Add opt-in configured and enforced power-limit fields: DCGM_FI_DEV_POWER_MGMT_LIMIT and DCGM_FI_DEV_ENFORCED_POWER_LIMIT.

  • Refresh the optional NVLink error-counter definitions.

  • Remove retired PCIe throughput fields from the supplied metric configurations. Custom configurations that use them fail and point to DCGM_FI_PROF_PCIE_TX_BYTES and DCGM_FI_PROF_PCIE_RX_BYTES.

  • Collect compute-instance-scoped DCGM fields from MIG compute instances and add the GPU_CI_ID label while preserving whole-GPU collection.

  • Attribute DRA MIG metrics to pods when the MIG device is visible to DCGM at startup, and refresh allocation metadata when ResourceSlices change.

  • Require an exporter restart after live Dynamic MIG creation or removal; changing the MIG layout while the exporter is running is not supported.

  • Add the optional pod_uid label to DRA-attributed metrics.

  • Reuse the kubelet pod-resources gRPC connection to prevent memory growth.

  • Prefer NVML MIG profile names and make MIG profile lookup available outside Kubernetes.

  • Deduplicate repeated job IDs within each HPC mapping file while preserving mappings across files.

  • Warn when HPC job mapping entries do not match discovered MIG instances.

Helm, containers, and packages#

  • Include all supported metric preset CSV files in binary packages.

  • Include third-party notices in archives, packages, and container images.

  • Include the dynamic linker cache in the distroless image so DCGM libraries can be found at startup.

  • Add Helm values for DaemonSet annotations.

  • Add Helm values for readiness and liveness probe timing and failure thresholds.

  • Render debug dump compression explicitly, including when it is disabled.

  • Document opt-in scheduling on nodes with the nvidia.com/gpu:NoSchedule taint.

  • Move raw Kubernetes deployment examples under deployment/examples.

  • Update the Helm chart and deployment examples for DCGM Exporter 4.8.4.

  • Update Go to 1.27.1 and refresh Go dependencies, container base images, Kubernetes tools, and development tooling.

Testing and documentation#

  • Expand CPU-only performance benchmarks for metric rendering and scraping.

  • Expand coverage for field-watch retention, oversized field groups, concurrent scrapes, DRA allocation, pre-existing MIG devices, and GPU bind/unbind lifecycle handling.

  • Simplify the repository README and direct users to the official DCGM Exporter installation, configuration, metrics, and release documentation.

4.6.0-4.8.3#

Released:

2026-07-15

Tested with DCGM:

4.6.0

Source:

4.6.0-4.8.3 on GitHub

Changelist#

Runtime and configuration#

  • Release DCGM Exporter 4.8.3 with DCGM 4.6.0 and go-dcgm v1.4601.1.

  • Improve go-dcgm policy, resource, field, and version API handling.

  • Add YAML configuration-file support.

  • Expand CLI and environment configuration.

  • Add configurable web read/write timeouts.

  • Add per-watch polling intervals.

  • Add combined device selectors.

  • Add VSOCK remote hostengine connections.

  • Use the local hostname for loopback remote hostengines.

  • Improve hot reload and last-known-good configuration handling.

  • Improve profiling and watcher recovery.

  • Improve startup, skipped-field, MIG, and remote-hostengine diagnostics.

  • Strengthen opt-in pprof and socket validation.

Metrics and Kubernetes#

  • Rename the Hostname label to hostname.

  • Add generic container-runtime label collection.

  • Support pod-label collection on MIG devices.

  • Allocate independent label maps for each entity.

  • Add DRA ResourceSlice v1 support.

  • Expand DRA v1 and MIG allocation handling.

  • Bound pod-resources responses to prevent resource-exhaustion errors.

  • Improve Kubernetes metadata diagnostics.

  • Add cumulative XID error totals.

  • Add cumulative clock-event totals.

  • Add GPU-health severity and category metadata.

  • Add Grace CPU serial labels.

  • Improve metric help text and metadata consistency.

  • Correct NVLink bandwidth metrics from counters to gauges.

  • Harden metric float parsing.

  • Improve field-status handling, metric exposition, and label sanitization.

  • Propagate HTTP request context through metric rendering.

Helm, containers, and packages#

  • Add Helm support for configuration files.

  • Add Helm image-digest configuration.

  • Update ServiceMonitor timing defaults.

  • Align RBAC with Kubernetes metadata and DRA features.

  • Make service-account token mounting configurable.

  • Send package-managed systemd logs to journald.

  • Correct container image provenance metadata.

  • Improve DCGM, NVPerf, and NVSDM library packaging.

  • Complete the Go 1.26.4 update.

  • Remove the direct dependency on the pkg/errors library.

  • Centralize release versions and consistency validation.

Testing and documentation#

  • Add a reusable E2E CLI and scenario framework.

  • Expand static, host, container, and Kubernetes test coverage.

  • Add repository-local agent-assisted development guidance.

  • Add the llms.txt metrics and deployment authoring contract.

  • Add Codex and Cursor project guidance.

4.5.3-4.8.2#

Released:

2026-05-07

Tested with DCGM:

4.5.3

Source:

4.5.3-4.8.2 on GitHub

  • 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.

  • Add per-process GPU metrics for time-sharing and MIG.

  • Make Helm priorityClassName configurable with explicit defaults.

  • Add MIG device support for HPC job labels.

  • 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.

4.5.2-4.8.1#

Released:

2026-02-09

Tested with DCGM:

4.5.2

Source:

4.5.2-4.8.1 on GitHub

  • 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

4.5.1-4.8.0#

Released:

2026-01-28

Tested with DCGM:

4.5.1

Source:

4.5.1-4.8.0 on GitHub

  • Update to DCGM 4.5.1

  • Enable beta monitoring of GPU bind/unbind events and automatic reloading.

  • Sync the default metric watchlist for Docker and Helm.

  • Fix health endpoint behavior.

  • Increase the default memory limit to 512 MiB.

  • Make scrapeTimeout configurable.

  • Fix P2P status mappings.

NOTICE: Helm chart now uses distroless container by default

4.4.2-4.7.1#

Released:

2025-12-10

Tested with DCGM:

4.4.2

Source:

4.4.2-4.7.1 on GitHub

  • Update Go-DCGM

  • Update XID error text for version 580.

  • Fix validation-time conversion from microseconds to nanoseconds.

  • Update the README with OpenObserve information and dashboards.

4.4.2-4.7.0#

Released:

2025-11-18

Tested with DCGM:

4.4.2

Source:

4.4.2-4.7.0 on GitHub

  • Improve startup and pipeline security.

  • Prevent a crash when the collector cannot initialize during startup validation.

  • Track unallocated GPUs in DRA.

  • Bound the label-cache size.

4.4.1-4.6.0#

Released:

2025-10-13

Tested with DCGM:

4.4.1

Source:

4.4.1-4.6.0 on GitHub

  • Add an allowlist for pod-label filtering.

  • Handle uninitialized maps.

  • Add the hostPID field to the Helm values file and DaemonSet template.

  • Add an option to fail when the NVML provider cannot initialize.

  • Improved support for GPU NvLink monitoring

4.4.1-4.5.2#

Released:

2025-09-17

Tested with DCGM:

4.4.1

Source:

4.4.1-4.5.2 on GitHub

  • Follow the FHS logging convention.

  • Enable metrics without pods by using kubernetes-enable-{dra,virtual-gpus}.

  • Add a flag to disable startup validation.

  • Reduce error logging when the HPC jobs directory does not exist.

4.4.0-4.5.0#

Released:

2025-08-19

Tested with DCGM:

4.4.0

Source:

4.4.0-4.5.0 on GitHub

  • Update to DCGM 4.4 and Cuda 13.0

  • Add Kubernetes UID support.

  • Create distroless container target

4.3.1-4.4.0#

Released:

2025-08-07

Tested with DCGM:

4.3.1

Source:

4.3.1-4.4.0 on GitHub

  • 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

4.2.3-4.2.0#

Released:

2025-07-11

Tested with DCGM:

4.2.3

Source:

4.2.3-4.2.0 on GitHub

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

4.2.3-4.1.3#

Released:

2025-06-05

Tested with DCGM:

4.2.3

Source:

4.2.3-4.1.3 on GitHub

  • Helm chart refactoring

  • Update podresourceapi to v1

  • Reload on config change

  • Support for internalTrafficePolicy

  • Fix hostname label in certain environments

4.2.3-4.1.1#

Released:

2025-05-15

Tested with DCGM:

4.2.3

Source:

4.2.3-4.1.1 on GitHub

  • Add log formatting

  • Workaround NvLink entity bug in DCGM

4.2.0-4.1.0#

Released:

2025-04-15

Tested with DCGM:

4.2.0

Source:

4.2.0-4.1.0 on GitHub

  • Update to DCGM 4.2.0

  • Update to latest Go-DCGM bindings

4.1.1-4.0.4#

Released:

2025-02-20

Tested with DCGM:

4.1.1

Source:

4.1.1-4.0.4 on GitHub

  • Include the DCGM proprietary package for DCP metrics on Ada and older GPUs.

4.1.1-4.0.3#

Released:

2025-02-19

Tested with DCGM:

4.1.1

Source:

4.1.1-4.0.3 on GitHub

  • Update to DCGM 4.1.1

  • Add support for GPU-sharing metrics in Kubernetes.

4.0.0-4.0.1#

Released:

2025-01-27

Tested with DCGM:

4.0.0

Source:

4.0.0-4.0.1 on GitHub

  • Update to DCGM 4.0.0

  • Major refactor to enable clean mock testing

  • Refactor metric collection to align with prometheus best practices

  • Many more bug fixes and improvements

3.3.9-3.6.1#

Released:

2024-11-19

Tested with DCGM:

3.3.9

Source:

3.3.9-3.6.1 on GitHub

  • Update to DCGM 3.3.9

  • Allow selecting the service’s ClusterIP - Remi

  • Configurable service monitor API value

3.3.8-3.6.0#

Released:

2024-09-19

Tested with DCGM:

3.3.8

Source:

3.3.8-3.6.0 on GitHub

  • Update DCGM to 3.3.8

  • [Helm] Enable custom metrics, mount ConfigMap by default (Chip Zoller)

3.3.7-3.5.0#

Released:

2024-07-25

Tested with DCGM:

3.3.7

Source:

3.3.7-3.5.0 on GitHub

Changes:

  • Make nvidia resource names configurable (lx1036)

  • Update default PCIe metrics name (koshieguchi)

  • Correct metric help text (pintohutch)

  • Add pci_bus_id label for metric (fungaren)

  • Update to DCGM 3.3.7

3.3.6-3.4.2#

Released:

2024-05-20

Tested with DCGM:

3.3.6

Source:

3.3.6-3.4.2 on GitHub

  • Enable HPC job ID as label with –hpc-job-mapping-dir

  • Add err_msg label for XID errors

  • Bug fixes, bump base container, etc

3.3.5-3.4.1#

Released:

2024-04-03

Tested with DCGM:

3.3.5

Source:

3.3.5-3.4.1 on GitHub

  • Fix for duplicate DCGM_FI_DEV_XID_ERRORS

  • Make kubelet pod-resources socket directory configurable

  • Allow setting runtimeClassName

  • use Linux container CPU quota

  • Update Go dependencies

  • Added DCGM logging options

  • many tests added

3.3.5-3.4.0#

Released:

2024-02-26

Tested with DCGM:

3.3.5

Source:

3.3.5-3.4.0 on GitHub

  • New DCGM_EXP_CLOCK_EVENTS_COUNT and DCGM_EXP_XID_ERRORS_COUNT metrics

  • Graceful handling of panics

  • Various lint fixes

  • Control modelName format

3.3.3-3.3.1#

Released:

2024-02-02

Tested with DCGM:

3.3.3

Source:

3.3.3-3.3.1 on GitHub

  • Fix for crash in dcgmGetCpuHierarchy

3.3.3-3.3.0#

Released:

2024-01-29

Tested with DCGM:

3.3.3

Source:

3.3.3-3.3.0 on GitHub

  • Update DCGM to 3.3.3

  • Enable Grace CPU support

  • Update Go to 1.21

  • TLS and auth support

  • Community fixes and enhancements (Thank you!)

3.3.0-3.2.0#

Released:

2023-11-08

Tested with DCGM:

3.3.0

Source:

3.3.0-3.2.0 on GitHub

  • Update DCGM to 3.3.0

  • Various community fixes and enhancements

3.2.5-3.1.7#

Released:

2023-08-31

Tested with DCGM:

3.2.5

Source:

3.2.5-3.1.7 on GitHub

  • Update DCGM to 3.2.5

  • Grafana fixes

3.1.8-3.1.5#

Released:

2023-06-01

Tested with DCGM:

3.1.8

Source:

3.1.8-3.1.5 on GitHub

  • Update DCGM to 3.1.8

3.1.7-3.1.4#

Released:

2023-03-31

Tested with DCGM:

3.1.7

Source:

3.1.7-3.1.4 on GitHub

  • Update DCGM to 3.1.7

  • Update Go to 1.18

  • Various library and community updates

3.1.6-3.1.3#

Released:

2023-02-09

Tested with DCGM:

3.1.6

Source:

3.1.6-3.1.3 on GitHub

  • Enable NODE_NAME env for metrics

  • Update to DCGM 3.1.6

  • Enable x-content-type-options in http header

  • Community enhancements

3.1.3-3.1.2#

Released:

2022-12-13

Tested with DCGM:

3.1.3

Source:

3.1.3-3.1.2 on GitHub

  • Update to DCGM 3.1.3

  • NvSwitch/NvLink metric support

  • Community bug fixes and enhancements

3.0.4-3.0.0#

Released:

2022-10-18

Tested with DCGM:

3.0.4

Source:

3.0.4-3.0.0 on GitHub

  • Update DCGM to 3.0.4

  • Update DCGM headers

2.4.7-2.6.11#

Released:

2022-09-28

Tested with DCGM:

2.4.7

Source:

2.4.7-2.6.11 on GitHub

Update DCGM to 2.4.7

2.4.6-2.6.10#

Released:

2022-08-09

Tested with DCGM:

2.4.6

Source:

2.4.6-2.6.10 on GitHub

  • “label” metric types

  • Skip unsupported metrics

2.4.6-2.6.8#

Released:

2022-07-18

Tested with DCGM:

2.4.6

Source:

2.4.6-2.6.8 on GitHub

  • Update DCGM to 2.4.6

  • Update to latest go-dcgm

2.4.5-2.6.7#

Released:

2022-06-07

Tested with DCGM:

2.4.5

Source:

2.4.5-2.6.7 on GitHub

  • Update DCGM to 2.4.5

  • Fixes for certain MIG metrics

2.3.6-2.6.6#

Released:

2022-05-23

Tested with DCGM:

2.3.6

Source:

2.3.6-2.6.6 on GitHub

  • Update DCGM to 2.3.6

  • Fix comment parsing in csv files

  • Some community patches

2.3.5-2.6.5#

Released:

2022-03-23

Tested with DCGM:

2.3.5

Source:

2.3.5-2.6.5 on GitHub

  • Update DCGM to 2.3.5

  • Fix setcap check in docker entrypoint

  • Update to latest go-dcgm bindings

2.3.4-2.6.4#

Released:

2022-02-28

Tested with DCGM:

2.3.4

Source:

2.3.4-2.6.4 on GitHub

  • Fixes for reporting MIG metrics

2.3.2-2.6.3#

Released:

2022-02-03

Tested with DCGM:

2.3.2

Source:

2.3.2-2.6.3 on GitHub

  • Update base image to 11.6.0

  • Enable multi-arch builds

2.3.2-2.6.2#

Released:

2022-01-20

Tested with DCGM:

2.3.2

Source:

2.3.2-2.6.2 on GitHub

Fixes for monitoring MIG devices

2.3.1-2.6.1#

Released:

2021-11-24

Tested with DCGM:

2.3.1

Source:

2.3.1-2.6.1 on GitHub

Use go-nvml for parsing MIG UUIDs.

2.3.1-2.6.0#

Released:

2021-10-28

Tested with DCGM:

2.3.1

Source:

2.3.1-2.6.0 on GitHub

  • Update DCGM to 2.3.1

  • Better MIG handling by default

  • Restore DCGM_FI_DEV_GPU_UTIL to default watchlist

2.2.9-2.5.0#

Released:

2021-09-13

Tested with DCGM:

2.2.9

Source:

2.2.9-2.5.0 on GitHub

Latest features and fixes, including ConfigMap support with “-m <namespace>:<name>”.