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/dockerfiles/Dockerfile.ollama

@@ -1,4 +1,4 @@
-FROM docker.io/ollama/ollama:0.9.6
+FROM docker.io/ollama/ollama:0.13.4
 
 ARG UID=1000
 ARG GID=1000

ansible/files/dockerfiles/Dockerfile.ollama.rocm

@@ -0,0 +1,12 @@
+FROM docker.io/ollama/ollama:0.13.4-rocm
+
+ARG UID=1000
+ARG GID=1000
+RUN groupadd --system --gid ${GID} worker && \
+    useradd --system --gid ${GID} --uid ${UID} --home /home/worker worker && \
+    mkdir -p /home/worker && chown worker:worker /home/worker
+
+WORKDIR /home/worker
+USER worker
+ENTRYPOINT ["/bin/ollama"]
+CMD ["serve"]

ansible/files/game_station/bin/system/gamescope-session

@@ -10,13 +10,14 @@ CONF_FILE="$HOME/.config/gamescope/gamescope.conf"
 RESOLUTION=${RESOLUTION:-1920x1080}
 WIDTH="${RESOLUTION%x*}"
 HEIGHT="${RESOLUTION#*x}"
+FRAMERATE=${FRAMERATE:-60}
 
 export __GL_GSYNC_ALLOWED=1
 export __GL_VRR_ALLOWED=1
 export __GL_SHADER_DISK_CACHE=1
 export __GL_SYNC_TO_VBLANK=0
-export LIBVA_DRIVER_NAME=nvidia
 export __GLX_VENDOR_LIBRARY_NAME=nvidia
+export LIBVA_DRIVER_NAME=nvidia
 export NVD_BACKEND=direct
 
 GAMESCOPE_BIN="$(command -v gamescope || true)"
@@ -45,6 +46,15 @@ if [[ -z $tmpdir || -z ${XDG_RUNTIME_DIR+x} ]]; then
 	exit 0
 fi
 
+# Setup MangoHud config file in temp directory
+export MANGOHUD_CONFIGFILE="${tmpdir:+$tmpdir/mangohud.config}"
+
+# Initially write no_display to our config file
+# so we don't get mangoapp showing up before Steam initializes
+# on OOBE and stuff.
+mkdir -p "$(dirname "$MANGOHUD_CONFIGFILE")"
+echo "no_display" > "$MANGOHUD_CONFIGFILE"
+
 # Chromium (and therefore Steam) ignore XCursor and use on the GTK config
 kwriteconfig6 --file gtk-3.0/settings.ini  --group Settings --key gtk-cursor-theme-name steam
 

ansible/files/game_station/bin/system/steam-launcher

@@ -21,7 +21,7 @@ export VKD3D_SWAPCHAIN_LATENCY_FRAMES=3
 # Let's try this across the board to see if it breaks anything
 # Helps performance in HZD, Cyberpunk, at least
 # Expose 8 physical cores, instead of 4c/8t
-export WINE_CPU_TOPOLOGY=8:0,1,2,3,4,5,6,7
+# export WINE_CPU_TOPOLOGY=8:0,1,2,3,4,5,6,7
 
 # To expose vram info from radv's patch we're including
 export WINEDLLOVERRIDES=dxgi=n

ansible/files/game_station/bin/system/steamos-select-branch

@@ -20,7 +20,7 @@ if [[ $# -eq 1 ]]; then
       exit 0
       ;;
     "stable" | "rc" | "beta" | "bc" | "preview" | "pc" | "main" | "staging")
-      log "Branch '$1' requested"
+      log "Branch '$1' requested"      
       exit 0
       ;;
   esac

ansible/files/game_station/systemd/user/session.slice

@@ -12,4 +12,5 @@ Description=User Core Session Slice
 Documentation=man:systemd.special(7)
 
 [Slice]
-CPUWeight=100
+CPUWeight=100
+# AllowedCPUs={{ general_cores }}

ansible/files/game_station/systemd/user/steam.slice

@@ -12,4 +12,5 @@ Description=User Core Session Slice
 Documentation=man:systemd.special(7)
 
 [Slice]
-CPUWeight=100
+CPUWeight=100
+# AllowedCPUs={{ general_cores }}

ansible/files/game_station/systemd/user/steamos.service

@@ -6,3 +6,4 @@ After=graphical-session.target
 ExecStart=/usr/local/bin/steamos
 EnvironmentFile=%t/steamos-environment
 Restart=on-failure
+# AllowedCPUs={{ general_cores }}

ansible/files/game_station/systemd/user/sunshine.slice

@@ -12,4 +12,5 @@ Description=User Core Session Slice
 Documentation=man:systemd.special(7)
 
 [Slice]
-CPUWeight=100
+CPUWeight=100
+# AllowedCPUs={{ sunshine_cores }}

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