feat: Add ROCm support and update Ollama configurations

- Created ansible playbooks for setting up AMD ROCm container environment.
- Added ROCm and AMD Graphics repositories in ansible tasks.
- Installed necessary ROCm packages including rocm-dkms and developer tools.
- Updated Dockerfile for Ollama to use version 0.13.4 and created a new Dockerfile for ROCm support.
- Enhanced game station scripts for better performance and added MangoHud configuration.
- Modified systemd service files to allow CPU core allocation.
- Updated playbooks for Ollama and Qwen2 to include new configurations and environment variables.
- Added Wake-on-LAN systemd service configuration.
- Updated Terraform configurations to include secret management for the web UI.
- Refactored Kubernetes deployment for open-webui to include secret handling and improved volume management.

ansible/files/scripts/auto_suspend.sh

@@ -22,7 +22,7 @@ if ! command -v nvidia-smi &> /dev/null; then
 else
     # Check 3: Is GPU idle?
     GPU_UTIL=$(nvidia-smi --query-gpu=utilization.gpu --format=csv,noheader,nounits | head -n 1)
-    GPU_THRESHOLD=10
+    GPU_THRESHOLD=0
 
     if (( GPU_UTIL > GPU_THRESHOLD )); then
         echo "GPU is busy (utilization=${GPU_UTIL}%). Skipping suspend."
@@ -37,5 +37,25 @@ else
     fi
 fi
 
+if | command -v amd-smi &> /dev/null; then
+    # Check 5: Is AMD GPU idle?
+    AMD_GPU_UTIL=$(amd-smi monitor --csv | head -n 2 | tail -n 1 | cut -d ',' -f 8)
+    AMD_GPU_THRESHOLD=0
+
+    if (( AMD_GPU_UTIL > AMD_GPU_THRESHOLD )); then
+        echo "AMD GPU is busy (utilization=${AMD_GPU_UTIL}%). Skipping suspend."
+        exit 0
+    fi
+
+    if amd-smi process --csv | head -n 2 | tail -n 1 | grep -q "No running processes detected"; then
+        echo "No processes running on AMD GPU."
+    else
+        echo "Processes are running on AMD GPU. Skipping suspend."
+        exit 0
+    fi
+else
+    echo "amd-smi not found, skipping AMD GPU check"
+fi
+
 echo "System idle. Suspending..."
 shutdown now