feat(rocm): ajoute rocm-smi (paquet rocm-smi-lib) sur gpu-01 (#85)

Monitoring GPU consolidé (util/VRAM/temp/clocks) façon nvidia-smi, aligné
sur la ROCm 7.2.4 installée (repo rocm, /opt/rocm/bin/rocm-smi déjà dans
le PATH). Appliqué sur gpu-01 : llama-server intact (GPU non perturbé).
Doc admin/ia/rocm.md + README du rôle mis à jour.

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
ALI YESILKAYA 2026-07-07 11:55:21 +02:00 committed by GitHub
parent 9c5b8d2048
commit 9b903c0311
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 28 additions and 5 deletions

View file

@ -6,7 +6,7 @@ Stack GPU AMD sur gpu-01 (AlmaLinux, RX 6700XT) : repos AMD, runtime ROCm, paque
- Clé GPG AMD + repos `amdgpu` et `rocm` (latest)
- kernel headers/devel (requis pour amdgpu-dkms)
- Runtime : rocminfo, rocm-hip-runtime, rocm-opencl-runtime, rocblas, hipblaslt
- Runtime : rocminfo, rocm-smi-lib (outil `rocm-smi`), rocm-hip-runtime, rocm-opencl-runtime, rocblas, hipblaslt
- Paquets build (pour le rôle `llama_server`) : hipblas-devel, rocblas-devel, hip-devel, hipcc, cmake, gcc-c++, patchelf
- `HSA_OVERRIDE_GFX_VERSION=10.3.0` **system-wide** (`/etc/environment`) et dans `profile.d`
- Ajoute le user ansible aux groupes `video` et `render`

View file

@ -13,6 +13,7 @@ rocm_gpg_key: "https://repo.radeon.com/rocm/rocm.gpg.key"
rocm_packages:
- rocminfo
- rocm-smi-lib # outil rocm-smi (monitoring GPU : util, VRAM, temp, conso, clocks)
- rocm-hip-runtime
- rocm-opencl-runtime
- rocblas