Nvidia Hotspot / VRAM temps (#583)

Thanks to @TimSirmovics
2021-10-27 16:34:07 +02:00
parent f157563c33
commit 1701eb91b2
2 changed files with 93 additions and 2 deletions
@@ -40,6 +40,9 @@ namespace LibreHardwareMonitor.Hardware.Gpu
        private readonly Sensor[] _powers;
        private readonly Sensor _powerUsage;
        private readonly Sensor[] _temperatures;
+        private uint _thermalSensorsMask;
+        private readonly Sensor _hotSpotTemperature;
+        private readonly Sensor _memoryJunctionTemperature;

        public NvidiaGpu(int adapterIndex, NvApi.NvPhysicalGpuHandle handle, NvApi.NvDisplayHandle? displayHandle, ISettings settings)
            : base(GetName(handle),
@@ -79,6 +82,32 @@ namespace LibreHardwareMonitor.Hardware.Gpu
                }
            }

+            // Thermal sensors.
+            _hotSpotTemperature = new Sensor("GPU Hot Spot", (int)thermalSettings.Count + 1, SensorType.Temperature, this, settings);
+            _memoryJunctionTemperature = new Sensor("GPU Memory Junction", (int)thermalSettings.Count + 2, SensorType.Temperature, this, settings);
+            bool hasAnyThermalSensor = false;
+
+            for (int thermalSensorsMaxBit = 0; thermalSensorsMaxBit < 32; thermalSensorsMaxBit++)
+            {
+                // Find the maximum thermal sensor mask value.
+                _thermalSensorsMask = 1u << thermalSensorsMaxBit;
+
+                GetThermalSensors(_thermalSensorsMask, out NvApi.NvStatus thermalSensorsStatus);
+                if (thermalSensorsStatus == NvApi.NvStatus.OK)
+                {
+                    hasAnyThermalSensor = true;
+                    continue;
+                }
+
+                _thermalSensorsMask--;
+                break;
+            }
+
+            if (!hasAnyThermalSensor)
+            {
+                _thermalSensorsMask = 0;
+            }
+
            // Clock frequencies.
            for (int clockVersion = 1; clockVersion <= 3; clockVersion++)
            {
@@ -434,6 +463,28 @@ namespace LibreHardwareMonitor.Hardware.Gpu
                }
            }

+            if (_thermalSensorsMask > 0)
+            {
+                NvApi.NvThermalSensors thermalSensors = GetThermalSensors(_thermalSensorsMask, out status);
+
+                if (status == NvApi.NvStatus.OK)
+                {
+                    _hotSpotTemperature.Value = thermalSensors.Temperatures[1] / 256.0f;
+                    _memoryJunctionTemperature.Value = thermalSensors.Temperatures[9] / 256.0f;
+                }
+
+                if (_hotSpotTemperature.Value != 0)
+                    ActivateSensor(_hotSpotTemperature);
+
+                if (_memoryJunctionTemperature.Value != 0)
+                    ActivateSensor(_memoryJunctionTemperature);
+            }
+            else
+            {
+                _hotSpotTemperature.Value = null;
+                _memoryJunctionTemperature.Value = null;
+            }
+
            if (_clocks is { Length: > 0 })
            {
                NvApi.NvGpuClockFrequencies clockFrequencies = GetClockFrequencies(out status);
@@ -928,6 +979,24 @@ namespace LibreHardwareMonitor.Hardware.Gpu
            return status == NvApi.NvStatus.OK ? settings : default;
        }

+        private NvApi.NvThermalSensors GetThermalSensors(uint mask, out NvApi.NvStatus status)
+        {
+            if (NvApi.NvAPI_GPU_ThermalGetSensors == null)
+            {
+                status = NvApi.NvStatus.Error;
+                return default;
+            }
+
+            var thermalSensors = new NvApi.NvThermalSensors()
+            {
+                Version = (uint)NvApi.MAKE_NVAPI_VERSION<NvApi.NvThermalSensors>(2),
+                Mask = mask
+            };
+
+            status = NvApi.NvAPI_GPU_ThermalGetSensors(_handle, ref thermalSensors);
+            return status == NvApi.NvStatus.OK ? thermalSensors : default;
+        }
+
        private NvApi.NvFanCoolersStatus GetFanCoolersStatus(out NvApi.NvStatus status)
        {
            if (NvApi.NvAPI_GPU_ClientFanCoolersGetStatus == null)
@@ -938,7 +1007,8 @@ namespace LibreHardwareMonitor.Hardware.Gpu

            var coolers = new NvApi.NvFanCoolersStatus
            {
-                Version = (uint)NvApi.MAKE_NVAPI_VERSION<NvApi.NvFanCoolersStatus>(1), Items = new NvApi.NvFanCoolersStatusItem[NvApi.MAX_FAN_COOLERS_STATUS_ITEMS]
+                Version = (uint)NvApi.MAKE_NVAPI_VERSION<NvApi.NvFanCoolersStatus>(1),
+                Items = new NvApi.NvFanCoolersStatusItem[NvApi.MAX_FAN_COOLERS_STATUS_ITEMS]
            };

            status = NvApi.NvAPI_GPU_ClientFanCoolersGetStatus(_handle, ref coolers);
@@ -983,7 +1053,8 @@ namespace LibreHardwareMonitor.Hardware.Gpu

            NvApi.NvDynamicPStatesInfo pStatesInfo = new()
            {
-                Version = (uint)NvApi.MAKE_NVAPI_VERSION<NvApi.NvDynamicPStatesInfo>(1), Utilizations = new NvApi.NvDynamicPState[NvApi.MAX_GPU_UTILIZATIONS]
+                Version = (uint)NvApi.MAKE_NVAPI_VERSION<NvApi.NvDynamicPStatesInfo>(1),
+                Utilizations = new NvApi.NvDynamicPState[NvApi.MAX_GPU_UTILIZATIONS]
            };

            status = NvApi.NvAPI_GPU_GetDynamicPstatesInfoEx(_handle, ref pStatesInfo);
@@ -17,6 +17,8 @@ namespace LibreHardwareMonitor.Interop
        public const int MAX_POWER_TOPOLOGIES = 4;
        public const int MAX_THERMAL_SENSORS_PER_GPU = 3;
        public const int MAX_USAGES_PER_GPU = 8;
+        public const int THERMAL_SENSOR_RESERVED_COUNT = 8;
+        public const int THERMAL_SENSOR_TEMPERATURE_COUNT = 32;

        public const int SHORT_STRING_MAX = 64;

@@ -39,6 +41,7 @@ namespace LibreHardwareMonitor.Interop
        public static readonly NvAPI_GPU_GetThermalSettingsDelegate NvAPI_GPU_GetThermalSettings;
        public static readonly NvAPI_GPU_GetUsagesDelegate NvAPI_GPU_GetUsages;
        public static readonly NvAPI_GPU_SetCoolerLevelsDelegate NvAPI_GPU_SetCoolerLevels;
+        public static readonly NvAPI_GPU_GetThermalSensorsDelegate NvAPI_GPU_ThermalGetSensors;

        private static readonly NvAPI_GetInterfaceVersionStringDelegate _nvAPI_GetInterfaceVersionString;
        private static readonly NvAPI_GPU_GetFullNameDelegate _nvAPI_GPU_GetFullName;
@@ -87,6 +90,7 @@ namespace LibreHardwareMonitor.Interop
                GetDelegate(0x814B209F, out NvAPI_GPU_ClientFanCoolersGetControl);
                GetDelegate(0xA58971A5, out NvAPI_GPU_ClientFanCoolersSetControl);
                GetDelegate(0x0EDCF624E, out NvAPI_GPU_ClientPowerTopologyGetStatus);
+                GetDelegate(0x65FE3AAD, out NvAPI_GPU_ThermalGetSensors);

                IsAvailable = true;
            }
@@ -152,6 +156,9 @@ namespace LibreHardwareMonitor.Interop
        [UnmanagedFunctionPointer(CallingConvention.Cdecl)]
        public delegate NvStatus NvAPI_GPU_SetCoolerLevelsDelegate(NvPhysicalGpuHandle gpuHandle, int coolerIndex, ref NvCoolerLevels NvCoolerLevels);

+        [UnmanagedFunctionPointer(CallingConvention.Cdecl)]
+        public delegate NvStatus NvAPI_GPU_GetThermalSensorsDelegate(NvPhysicalGpuHandle gpuHandle, ref NvThermalSensors nvThermalSensors);
+
        public enum NvFanControlMode : uint
        {
            Auto = 0,
@@ -502,6 +509,19 @@ namespace LibreHardwareMonitor.Interop
            public NvLevel[] Levels;
        }

+        [StructLayout(LayoutKind.Sequential, Pack = 8)]
+        public struct NvThermalSensors
+        {
+            internal uint Version;
+            internal uint Mask;
+
+            [MarshalAs(UnmanagedType.ByValArray, SizeConst = THERMAL_SENSOR_RESERVED_COUNT)]
+            internal int[] Reserved;
+
+            [MarshalAs(UnmanagedType.ByValArray, SizeConst = THERMAL_SENSOR_TEMPERATURE_COUNT)]
+            internal int[] Temperatures;
+        }
+
        [StructLayout(LayoutKind.Sequential, Pack = 8)]
        internal struct NvMemoryInfo
        {