feat: GPU-Auto-Erkennung (nvidia_smi, rocm-smi); Testserver-Aufraeumen dokumentiert

scripts/gpu-rocm.sh

@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+# gpu-rocm.sh — AMD-GPU-Metriken via rocm-smi im Influx-Format.
+# Wird vom Telegraf-exec-Input gpu.conf aufgerufen (nur wenn rocm-smi existiert).
+set -eu
+command -v rocm-smi >/dev/null 2>&1 || exit 0
+
+# rocm-smi --showuse --showtemp --showpower --showmeminfo vram --csv
+rocm-smi --showuse --showtemp --showpower --showmeminfo vram --csv 2>/dev/null \
+| awk -F, '
+  NR == 1 { next }
+  {
+    gsub(/^ +| +$/, "", $1); gsub(/^ +| +$/, "", $2); gsub(/^ +| +$/, "", $3); gsub(/^ +| +$/, "", $4);
+    card = $1
+    for (i = 2; i <= NF; i++) {
+      if ($i ~ /%/)      { g = $i; gsub(/%/, "", g); printf "gpu,gpu=%s,metric=utilization_gpu value=%s\n", card, g }
+      if ($i ~ /c$/)     { t = $i; gsub(/c/, "", t); printf "gpu,gpu=%s,metric=temperature_gpu value=%s\n", card, t }
+      if ($i ~ /W$/)     { p = $i; gsub(/W/, "", p); printf "gpu,gpu=%s,metric=power_draw value=%s\n", card, p }
+      if ($i ~ /^[0-9]+$/) { printf "gpu,gpu=%s,metric=memory_used value=%s\n", card, $i }
+    }
+  }' 2>/dev/null || true