From e0b67dabcfc8223463c92e157cc4b7b169f4d8f9 Mon Sep 17 00:00:00 2001 From: Yan Sun Date: Wed, 8 Jul 2026 10:53:43 -0700 Subject: [PATCH 1/2] docs: add MI350P, R9700S, R9600D, and W7900 to supported hardware table (#1582) * docs: add MI350P and Radeon AI Pro to supported GPUs table in docs/index.md The Supported Hardware table in docs/index.md was missing two GPU families that are actively validated in the platform-release CI pipeline and documented in the v1.5.1-beta release notes: - AMD Instinct MI350P (CDNA4/gfx950): has dedicated CI job, pytest fixtures, amdgpu-features.json entry, and Kubernetes+OpenShift support since v1.5.1-beta. - AMD Radeon AI Pro R9700S (RDNA4/gfx1201) and W7900 (RDNA3/gfx1100): have platform-release CI jobs, amdgpu-features.json entries, and Kubernetes support since v1.5.1-beta. Also adds MI308X (CDNA3/gfx942), which has SR-IOV exporter CI coverage alongside MI300X and MI350X. Expands the table into two sub-sections (Instinct / Radeon AI Pro) and adds an Architecture column for clarity. Includes a note on Radeon feature limits. Plan: docs-internal/knowledge/plans/2026-07-06-supported-gpus-update.md Co-Authored-By: Claude * docs: add GPU Operator version column to supported hardware table Maps each GPU model to the GPU Operator version that first introduced support, derived from release notes: - MI210/MI250/MI300X: v1.0.0+ - MI325X/MI350X/MI355X: v1.4.0+ - MI350P/R9700S/R9600D/W7900: v1.5.1+ Co-Authored-By: Claude * docs: remove internal IP from plan file Co-Authored-By: Claude * docs: remove GPU Operator version column from supported hardware table Keep the table two-column; the per-GPU operator-version mapping added noise without clear long-term maintenance value. Co-Authored-By: Claude --------- Co-authored-by: Claude (cherry picked from commit c9e6c93375d79bbde713a480cf84a9c6ff639f5c) --- .../plans/2026-07-06-supported-gpus-update.md | 61 +++++++++++++++++++ docs/index.md | 4 ++ 2 files changed, 65 insertions(+) create mode 100644 docs-internal/knowledge/plans/2026-07-06-supported-gpus-update.md diff --git a/docs-internal/knowledge/plans/2026-07-06-supported-gpus-update.md b/docs-internal/knowledge/plans/2026-07-06-supported-gpus-update.md new file mode 100644 index 00000000..de0494f6 --- /dev/null +++ b/docs-internal/knowledge/plans/2026-07-06-supported-gpus-update.md @@ -0,0 +1,61 @@ +# Supported GPUs Documentation Update: Add MI350P and Radeon AI PRO GPUs + +- **Date:** 2026-07-06 +- **Author:** Claude Code agent +- **Related PR(s):** feat/supported-gpus-docs +- **Related issue(s) / JIRA:** N/A + +## Context + +The `docs/index.md` Supported Hardware table listed only AMD Instinct MI210, +MI250, MI300X, MI325X, MI350X, and MI355X. Several GPU models that have been +actively validated in the CI pipeline and test infrastructure were missing: + +**MI350P** — CDNA4 GPU: +- Dedicated platform-release CI job in `tests/jobs/platform-release/mi350p/` +- Pytest fixtures in `test_config_manager.py` and `test_dra_partitioning.py` +- Entry in `amdgpu-features.json` and `metrics-support.json` +- v1.5.1-beta release notes explicitly call out MI350P on both K8s and OpenShift + +**Radeon AI PRO R9700S** — RDNA4 GPU: +- Active platform-release CI job (`tests/jobs/platform-release/radeon/`, gpu-series=R9700S) +- Entry in `amdgpu-features.json` and `metrics-support.json` +- Added in commit `9c5ef17e` (R9700 NFD rules) and subsequent Radeon fix commits + +**Radeon AI PRO R9600D** — RDNA4 GPU: +- Dedicated test support added in commit `9ecda4ee` +- Profiler metrics support confirmed in commit `e695467e` +- Tested on physical hardware (internal lab node, 8x R9600D) + +**Radeon Pro W7900** — RDNA3 GPU: +- Active platform-release CI jobs (K8s and OpenShift, gpu-series=W7900) +- Entry in `amdgpu-features.json` +- Profiler metrics support confirmed in commit `bdcde161` + +## Approach + +Add MI350P, R9700S, R9600D, and W7900 to the existing `### Supported Hardware` +table in `docs/index.md`, preserving the original two-column format. Instinct +GPUs remain at the top in version-descending order; Radeon AI PRO GPUs follow. + +A follow-up commit temporarily added a third **GPU Operator Version** column +mapping each GPU to the operator version that first introduced support. That +column was removed after review to keep the table minimal — the final table +stays two-column (`GPU`, `Support status`). + +## Scope + +- **In scope:** `docs/index.md` Supported Hardware table only. +- **Out of scope:** Architecture columns, OS/Platform matrix, release notes, + per-GPU operator-version column, any code changes. + +## Validation + +- Visual review of the rendered Markdown table. +- Cross-check each GPU against `tests/pytests/lib/files/amdgpu-features.json` + and platform-release CI job directories. + +## Risks and rollback + +- **Risk:** Minimal — purely additive documentation change. +- **Rollback:** Remove the added rows from `docs/index.md`. No code changes involved. diff --git a/docs/index.md b/docs/index.md index 2dd9b7b0..efac65c6 100644 --- a/docs/index.md +++ b/docs/index.md @@ -21,10 +21,14 @@ The AMD GPU Operator simplifies the deployment and management of AMD Instinct GP | --- | --- | | AMD Instinct™ MI355X | ✅ Supported | | AMD Instinct™ MI350X | ✅ Supported | +| AMD Instinct™ MI350P | ✅ Supported | | AMD Instinct™ MI325X | ✅ Supported | | AMD Instinct™ MI300X | ✅ Supported | | AMD Instinct™ MI250 | ✅ Supported | | AMD Instinct™ MI210 | ✅ Supported | +| AMD Radeon™ AI PRO R9700S | ✅ Supported | +| AMD Radeon™ AI PRO R9600D | ✅ Supported | +| AMD Radeon™ Pro W7900 | ✅ Supported | ### OS & Platform Support Matrix From b0dbcf322d674d02839a664ce2de54ca8279b2c0 Mon Sep 17 00:00:00 2001 From: Praveen Kumar Shanmugam <58961022+spraveenio@users.noreply.github.com> Date: Wed, 8 Jul 2026 12:50:43 -0700 Subject: [PATCH 2/2] Delete docs-internal/knowledge/plans/2026-07-06-supported-gpus-update.md --- .../plans/2026-07-06-supported-gpus-update.md | 61 ------------------- 1 file changed, 61 deletions(-) delete mode 100644 docs-internal/knowledge/plans/2026-07-06-supported-gpus-update.md diff --git a/docs-internal/knowledge/plans/2026-07-06-supported-gpus-update.md b/docs-internal/knowledge/plans/2026-07-06-supported-gpus-update.md deleted file mode 100644 index de0494f6..00000000 --- a/docs-internal/knowledge/plans/2026-07-06-supported-gpus-update.md +++ /dev/null @@ -1,61 +0,0 @@ -# Supported GPUs Documentation Update: Add MI350P and Radeon AI PRO GPUs - -- **Date:** 2026-07-06 -- **Author:** Claude Code agent -- **Related PR(s):** feat/supported-gpus-docs -- **Related issue(s) / JIRA:** N/A - -## Context - -The `docs/index.md` Supported Hardware table listed only AMD Instinct MI210, -MI250, MI300X, MI325X, MI350X, and MI355X. Several GPU models that have been -actively validated in the CI pipeline and test infrastructure were missing: - -**MI350P** — CDNA4 GPU: -- Dedicated platform-release CI job in `tests/jobs/platform-release/mi350p/` -- Pytest fixtures in `test_config_manager.py` and `test_dra_partitioning.py` -- Entry in `amdgpu-features.json` and `metrics-support.json` -- v1.5.1-beta release notes explicitly call out MI350P on both K8s and OpenShift - -**Radeon AI PRO R9700S** — RDNA4 GPU: -- Active platform-release CI job (`tests/jobs/platform-release/radeon/`, gpu-series=R9700S) -- Entry in `amdgpu-features.json` and `metrics-support.json` -- Added in commit `9c5ef17e` (R9700 NFD rules) and subsequent Radeon fix commits - -**Radeon AI PRO R9600D** — RDNA4 GPU: -- Dedicated test support added in commit `9ecda4ee` -- Profiler metrics support confirmed in commit `e695467e` -- Tested on physical hardware (internal lab node, 8x R9600D) - -**Radeon Pro W7900** — RDNA3 GPU: -- Active platform-release CI jobs (K8s and OpenShift, gpu-series=W7900) -- Entry in `amdgpu-features.json` -- Profiler metrics support confirmed in commit `bdcde161` - -## Approach - -Add MI350P, R9700S, R9600D, and W7900 to the existing `### Supported Hardware` -table in `docs/index.md`, preserving the original two-column format. Instinct -GPUs remain at the top in version-descending order; Radeon AI PRO GPUs follow. - -A follow-up commit temporarily added a third **GPU Operator Version** column -mapping each GPU to the operator version that first introduced support. That -column was removed after review to keep the table minimal — the final table -stays two-column (`GPU`, `Support status`). - -## Scope - -- **In scope:** `docs/index.md` Supported Hardware table only. -- **Out of scope:** Architecture columns, OS/Platform matrix, release notes, - per-GPU operator-version column, any code changes. - -## Validation - -- Visual review of the rendered Markdown table. -- Cross-check each GPU against `tests/pytests/lib/files/amdgpu-features.json` - and platform-release CI job directories. - -## Risks and rollback - -- **Risk:** Minimal — purely additive documentation change. -- **Rollback:** Remove the added rows from `docs/index.md`. No code changes involved.