Systemd job to set card power limit

throttle_powerlimit/throttle_instinct.sh

@@ -0,0 +1,45 @@
+#!/bin/bash
+
+watt_limit=120
+
+for i in {0..10}; do
+    card="/sys/class/drm/card$i"
+    # Skip non-existing, virtual or non-GPU devices
+    if [ ! -d "$card" ] || [ ! -e "$card/device" ]; then
+        continue
+    fi
+
+    # Resolve the PCI device path
+    device_path=$(readlink -f "$card/device")
+
+    # Read vendor and device IDs from sysfs
+    vendor_id=$(cat "$device_path/vendor")
+    device_id=$(cat "$device_path/device")
+
+    if [ "$vendor_id" == "0x1002" ] && [ "$device_id" == "0x66a0" ]; then 
+	    echo "Vega detected, limiting to $watt_limit W."
+
+	    # TODO check if the hwmon / power indices are stable with multiple cards or 
+	    # make the script robust to always set the correct power cap
+	    found=0
+	    for hwmon in /sys/class/hwmon/hwmon*; do
+        	if [ -e "$hwmon/device" ]; then
+                	hwmon_dev=$(readlink -f "$hwmon/device")
+                	if [ "$hwmon_dev" = "$device_path" ]; then
+                    		power_cap_file="$hwmon/power1_cap"
+                    		if [ -e "$power_cap_file" ]; then
+                        		echo ${watt_limit}000000 > $power_cap_file
+                        		found=1
+                        		break
+                    		fi
+                	fi
+            	fi
+	    done
+
+            if [ $found -eq 0 ]; then
+               echo "Warning: Could not find power1_cap for this GPU."
+            fi
+    fi 
+
+    echo
+done