mirror of
https://github.com/Alkatrazz24/Funk-lab.git
synced 2026-07-08 13:54:41 +02:00
docs(admin): add ops documentation and fix hermes dashboard service
- Remove admin/ from .gitignore (repo is now private) - Add --tui flag to hermes-dashboard.service template - Document hermes-tui wrapper, dashboard access, TUI usage Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
parent
1544a2fc6d
commit
dca71144d6
17 changed files with 1979 additions and 4 deletions
116
admin/rocm.md
Normal file
116
admin/rocm.md
Normal file
|
|
@ -0,0 +1,116 @@
|
|||
# ROCm — GPU AMD RX 6700XT (gpu-01)
|
||||
|
||||
ROCm tourne directement sur l'hôte AlmaLinux (pas dans des conteneurs).
|
||||
La RX 6700XT (gfx1031) n'est pas officiellement supportée → `HSA_OVERRIDE_GFX_VERSION=10.3.0` obligatoire.
|
||||
|
||||
---
|
||||
|
||||
## Vérification de base
|
||||
|
||||
```bash
|
||||
# Le GPU est-il détecté par le kernel ?
|
||||
ssh g01 "lspci | grep -i amd | grep -i vga"
|
||||
|
||||
# Le module amdgpu est-il chargé ?
|
||||
ssh g01 "lsmod | grep amdgpu"
|
||||
|
||||
# ROCm voit-il le GPU ? (doit lister la RX 6700XT)
|
||||
ssh g01 "HSA_OVERRIDE_GFX_VERSION=10.3.0 rocminfo | grep -A5 'Agent 2'"
|
||||
|
||||
# Version ROCm installée
|
||||
ssh g01 "rocminfo | grep 'ROCk'"
|
||||
|
||||
# GPU info complet
|
||||
ssh g01 "HSA_OVERRIDE_GFX_VERSION=10.3.0 rocminfo"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Variable HSA_OVERRIDE
|
||||
|
||||
```bash
|
||||
# Vérifier qu'elle est bien définie system-wide
|
||||
ssh g01 "cat /etc/environment | grep HSA"
|
||||
|
||||
# Vérifier dans profile.d (sessions interactives)
|
||||
ssh g01 "cat /etc/profile.d/rocm.sh"
|
||||
|
||||
# Tester manuellement (si la variable n'est pas dans l'env courant)
|
||||
ssh g01 "HSA_OVERRIDE_GFX_VERSION=10.3.0 rocminfo | grep gfx"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Modules kernel
|
||||
|
||||
```bash
|
||||
# Statut du module amdgpu
|
||||
ssh g01 "lsmod | grep amdgpu"
|
||||
|
||||
# Recharger le module (si nécessaire)
|
||||
ssh g01 "sudo modprobe -r amdgpu && sudo modprobe amdgpu"
|
||||
|
||||
# Logs kernel liés au GPU
|
||||
ssh g01 "sudo dmesg | grep -i amdgpu | tail -20"
|
||||
ssh g01 "sudo dmesg | grep -i 'drm\|gpu' | tail -20"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Repos AMD
|
||||
|
||||
```bash
|
||||
# Vérifier les repos installés
|
||||
ssh g01 "sudo dnf repolist | grep -i 'amd\|rocm'"
|
||||
|
||||
# Voir les packages ROCm installés
|
||||
ssh g01 "sudo dnf list installed | grep -i rocm"
|
||||
ssh g01 "sudo rpm -qa | grep rocm"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Groupes utilisateur (accès GPU)
|
||||
|
||||
```bash
|
||||
# L'user doit être dans video et render
|
||||
ssh g01 "groups ansible"
|
||||
|
||||
# Si absent — ajouter manuellement
|
||||
ssh g01 "sudo usermod -aG video,render ansible"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Debug ROCm / LM Studio
|
||||
|
||||
```bash
|
||||
# Tester OpenCL
|
||||
ssh g01 "HSA_OVERRIDE_GFX_VERSION=10.3.0 clinfo | head -30"
|
||||
|
||||
# Voir l'utilisation GPU en temps réel
|
||||
ssh g01 "watch -n 1 'cat /sys/class/drm/card*/device/gpu_busy_percent'"
|
||||
|
||||
# Température GPU
|
||||
ssh g01 "cat /sys/class/drm/card*/device/hwmon/hwmon*/temp1_input"
|
||||
# (valeur en millidegrés → diviser par 1000)
|
||||
|
||||
# VRAM utilisée
|
||||
ssh g01 "cat /sys/class/drm/card*/device/mem_info_vram_used"
|
||||
ssh g01 "cat /sys/class/drm/card*/device/mem_info_vram_total"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Fallback Vulkan (si ROCm ne fonctionne pas)
|
||||
|
||||
Si `rocminfo` ne liste pas le GPU malgré `HSA_OVERRIDE`, LM Studio peut utiliser Vulkan :
|
||||
|
||||
```bash
|
||||
# Vérifier que Vulkan fonctionne
|
||||
ssh g01 "vulkaninfo 2>/dev/null | grep 'GPU id'"
|
||||
|
||||
# Installer les outils Vulkan si absents
|
||||
ssh g01 "sudo dnf install -y vulkan-tools"
|
||||
```
|
||||
Dans LM Studio : Settings → GPU → Backend → Vulkan
|
||||
Loading…
Add table
Add a link
Reference in a new issue