fix(2.0): GPU-Metrik findet Strix-Halo-Card dynamisch (card1) + GTT

Real-Box-Befund: GPU ist card1 (nicht hardcoded card0), Connector-Dirs
uebersprungen. Zusaetzlich GTT (echter Unified-Memory-Pool) statt nur
VRAM-Carve-out. Frontend zeigt GTT bevorzugt.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/services/system.py

@@ -23,16 +23,25 @@ def _read_int(path: str) -> int | None:
 
 
 def _gpu_sysfs() -> dict | None:
-    """AMD-GPU-Auslastung + VRAM via sysfs (Linux). None, wenn nicht vorhanden."""
-    base = "/sys/class/drm/card0/device"
-    if not os.path.isdir(base):
-        return None
-    busy = _read_int(f"{base}/gpu_busy_percent")
-    vram_used = _read_int(f"{base}/mem_info_vram_used")
-    vram_total = _read_int(f"{base}/mem_info_vram_total")
-    if busy is None and vram_used is None:
-        return None
-    return {"busy_percent": busy, "vram_used": vram_used, "vram_total": vram_total}
+    """AMD-GPU-Auslastung + Speicher via sysfs (Linux). Findet die Basis-Card
+    dynamisch (Strix Halo ist oft card1, nicht card0) und überspringt die
+    Connector-Verzeichnisse (card1-DP-1 …). Strix Halo nutzt Unified Memory →
+    GTT ist der eigentliche große Pool; VRAM ist nur der kleine Carve-out."""
+    for dev in sorted(glob.glob("/sys/class/drm/card*/device")):
+        card = dev.split("/")[-2]            # z.B. "card1" oder "card1-DP-1"
+        if "-" in card:                      # Connector-Dir → kein GPU-Device
+            continue
+        busy = _read_int(f"{dev}/gpu_busy_percent")
+        if busy is None:
+            continue
+        return {
+            "busy_percent": busy,
+            "vram_used": _read_int(f"{dev}/mem_info_vram_used"),
+            "vram_total": _read_int(f"{dev}/mem_info_vram_total"),
+            "gtt_used": _read_int(f"{dev}/mem_info_gtt_used"),
+            "gtt_total": _read_int(f"{dev}/mem_info_gtt_total"),
+        }
+    return None
 
 
 def _temps() -> dict | None: