From 89fe168656af46ea752a19840d5d96ecb2bc64c2 Mon Sep 17 00:00:00 2001 From: ci-penbot-01 Date: Wed, 22 Jul 2026 03:33:25 +0530 Subject: [PATCH 1/2] [ROCM-26020] docs: move Radeon gfx_activity idle uptick to Fixes (v1.5.1) (#1621) (#1623) The bundled Device Metrics Exporter fixes the Radeon AI idle gfx cold-read (gpuagent cold-read sample drop on resume). Move the release note from Known Limitations to Fixes in the v1.5.1 section. (cherry picked from commit 16a9f2b5f2370d5b1d45dc8d8331bb11283e257e) Co-authored-by: Bhanu Kiran Atturu (cherry picked from commit 0686dbdcbd3dd9b8f45bc7e2b1396a2347d363ac) --- ...6-07-21-rocm-26020-gfx-idle-releasenote.md | 35 +++++++++++++++++++ docs/releasenotes.md | 4 +-- 2 files changed, 36 insertions(+), 3 deletions(-) create mode 100644 docs-internal/knowledge/plans/2026-07-21-rocm-26020-gfx-idle-releasenote.md diff --git a/docs-internal/knowledge/plans/2026-07-21-rocm-26020-gfx-idle-releasenote.md b/docs-internal/knowledge/plans/2026-07-21-rocm-26020-gfx-idle-releasenote.md new file mode 100644 index 000000000..e28813c2c --- /dev/null +++ b/docs-internal/knowledge/plans/2026-07-21-rocm-26020-gfx-idle-releasenote.md @@ -0,0 +1,35 @@ +# ROCM-26020: move Radeon gfx_activity idle uptick to Fixes (v1.5.1) + +- **Date:** 2026-07-21 +- **Author:** Bhanu Kiran Atturu +- **Related PR(s):** pensando/gpu-operator#1621 +- **Related issue(s) / JIRA:** ROCM-26020 + +## Context + +The v1.5.1 release notes listed the Radeon AI `gfx_activity` idle uptick under +Known Limitations. The bundled Device Metrics Exporter now fixes it (gpuagent +Navi48 idle gfx cold-read fix, DME commit 9f8bc9e3c), so the entry belongs under +Fixes. + +## Approach + +- `docs/releasenotes.md`: move the `gfx_activity` idle-uptick entry from the + v1.5.1 Known Limitations section to the v1.5.1 Fixes section, reworded to state + that `gfx_activity` now shows true idle values (per review feedback). + +## Scope + +- **In scope:** operator v1.5.1 release-notes wording. +- **Out of scope:** the DME-side release note (tracked in the DME repo) and the + underlying gpuagent fix. + +## Validation + +- Documentation only; no build. Wording confirmed against reviewer feedback on + the PR. + +## Risks and rollback + +- Known risks: none (docs only). +- Rollback: revert the commit. diff --git a/docs/releasenotes.md b/docs/releasenotes.md index 66cf08551..261449683 100644 --- a/docs/releasenotes.md +++ b/docs/releasenotes.md @@ -38,6 +38,7 @@ The AMD GPU Operator v1.5.1 release extends hardware support to the **AMD Instin - Platform support additions for MI350P and Radeon AI PRO GPU families. - **Driver upgrade reboot could leave operands with `exec format error`** — the driver-upgrade reboot pod now flushes filesystems (`sync`) before rebooting the node, and the DRA driver is now included in the pre-reboot drain list, preventing truncated operand executables after an operator-managed upgrade reboot. [[ROCm/gpu-operator#606]](https://github.com/ROCm/gpu-operator/pull/606) - **DRA driver requires v1.0.1+ with amdgpu driver 6.19.14 / release 31.40 and later** — starting with amdgpu driver `6.19.14` (packaged release `31.40`), a change in the driver's version-reporting scheme prevented earlier DRA driver versions from advertising GPUs, so DRA workloads could not be scheduled. When using the DRA driver with amdgpu `6.19.14` / `31.40` or later, deploy DRA driver **v1.0.1 or later**. +- **Device Metrics Exporter: `gfx_activity` idle uptick on Radeon AI resolved** — on Radeon AI platforms, `gfx_activity` now shows true idle values. ### Known Limitations @@ -50,9 +51,6 @@ The AMD GPU Operator v1.5.1 release extends hardware support to the **AMD Instin - **Node Labeller reports incorrect product name for Radeon AI GPUs** - On Radeon AI platforms, the `amd.com/gpu.product-name` node label is populated as `AMD_Radeon_Graphics` instead of the specific model name (e.g., `AMD Radeon AI Pro R9700S`). This is a known issue and will be fixed in a future release. -- **Device Metrics Exporter: `gfx_activity` idle uptick on Radeon AI** - - On Radeon AI platforms, `gfx_activity` may show a spurious uptick when the GPU is idle. This is a cosmetic issue and does not indicate actual GPU activity. - ## GPU Operator v1.5.0 Release Notes The AMD GPU Operator v1.5.0 release introduces support for **Dynamic Resource Allocation (DRA)** as an alternative to the device plugin, **Auto Node Remediation (ANR)** for automated recovery of unhealthy GPU worker nodes, and **Node Problem Detector (NPD)** integration for surfacing GPU-related node conditions. The release also brings broader configurability across the operator (KMM, kubelet socket path, custom package repositories, host network configs, global image pull secrets) and updates the managed stack to ROCm 7.2.1. From a607ade5cfebe117950999aeb40af6fbd2598312 Mon Sep 17 00:00:00 2001 From: Praveen Kumar Shanmugam <58961022+spraveenio@users.noreply.github.com> Date: Tue, 21 Jul 2026 15:19:14 -0700 Subject: [PATCH 2/2] Delete docs-internal/knowledge/plans/2026-07-21-rocm-26020-gfx-idle-releasenote.md --- ...6-07-21-rocm-26020-gfx-idle-releasenote.md | 35 ------------------- 1 file changed, 35 deletions(-) delete mode 100644 docs-internal/knowledge/plans/2026-07-21-rocm-26020-gfx-idle-releasenote.md diff --git a/docs-internal/knowledge/plans/2026-07-21-rocm-26020-gfx-idle-releasenote.md b/docs-internal/knowledge/plans/2026-07-21-rocm-26020-gfx-idle-releasenote.md deleted file mode 100644 index e28813c2c..000000000 --- a/docs-internal/knowledge/plans/2026-07-21-rocm-26020-gfx-idle-releasenote.md +++ /dev/null @@ -1,35 +0,0 @@ -# ROCM-26020: move Radeon gfx_activity idle uptick to Fixes (v1.5.1) - -- **Date:** 2026-07-21 -- **Author:** Bhanu Kiran Atturu -- **Related PR(s):** pensando/gpu-operator#1621 -- **Related issue(s) / JIRA:** ROCM-26020 - -## Context - -The v1.5.1 release notes listed the Radeon AI `gfx_activity` idle uptick under -Known Limitations. The bundled Device Metrics Exporter now fixes it (gpuagent -Navi48 idle gfx cold-read fix, DME commit 9f8bc9e3c), so the entry belongs under -Fixes. - -## Approach - -- `docs/releasenotes.md`: move the `gfx_activity` idle-uptick entry from the - v1.5.1 Known Limitations section to the v1.5.1 Fixes section, reworded to state - that `gfx_activity` now shows true idle values (per review feedback). - -## Scope - -- **In scope:** operator v1.5.1 release-notes wording. -- **Out of scope:** the DME-side release note (tracked in the DME repo) and the - underlying gpuagent fix. - -## Validation - -- Documentation only; no build. Wording confirmed against reviewer feedback on - the PR. - -## Risks and rollback - -- Known risks: none (docs only). -- Rollback: revert the commit.