#!/bin/bash # GPU deep state check echo "=== dmesg (sudo) - GPU related ===" sudo dmesg | grep -iE "amdgpu|gfx|gpu|drm|kfd|reset|fault|error|hang|gfxoff|power" | tail -60 echo "" echo "=== GPU power state ===" cat /sys/class/drm/card*/device/power_dpm_force_performance_level 2>/dev/null echo "" cat /sys/class/drm/card*/device/pp_dpm_sclk 2>/dev/null echo "" cat /sys/class/drm/card*/device/pp_dpm_mclk 2>/dev/null echo "" echo "=== GPU runtime status ===" cat /sys/class/drm/card*/device/power/runtime_status 2>/dev/null echo "" echo "=== GFXOFF ===" # Check if GFXOFF is disabled cat /sys/module/amdgpu/parameters/noretry 2>/dev/null cat /sys/kernel/debug/dri/*/amdgpu_gfxoff_status 2>/dev/null 2>&1 echo "" echo "=== Force GPU wake ===" sudo bash -c 'echo high > /sys/class/drm/card1/device/power_dpm_force_performance_level 2>/dev/null || echo "card1 failed"' sudo bash -c 'echo high > /sys/class/drm/card0/device/power_dpm_force_performance_level 2>/dev/null || echo "card0 failed"' sleep 1 echo "After wake:" cat /sys/class/drm/card*/device/pp_dpm_sclk 2>/dev/null cat /sys/class/drm/card*/device/power/runtime_status 2>/dev/null echo "" echo "=== rocm-smi after wake ===" rocm-smi 2>&1 | grep -E "SCLK|MCLK|GPU%|Perf|Power"