public static DeviceQuery QueryGpuStatus(string fullPciBusID)
{
var errorMessage = string.Empty;
if (!Initialized)
{
return(null);
}
if (!GetDeviceByPciBusID(fullPciBusID, out NvmlDevice device, out errorMessage))
{
return(null);
}
if (!string.IsNullOrWhiteSpace(errorMessage))
{
CUDA.Log(Level.Error, $"API failed to get device by PCI bus ID ({fullPciBusID}): {errorMessage}");
return(null);
}
var utilization = new NvmlUtilization();
uint fanSpeed = 0u, temperature = 0u, power = 0u, gpuClock = 0u, vramClock = 0u;
var response =
RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? PInvokeWindows.NvmlDeviceGetFanSpeed(device, ref fanSpeed)
: RuntimeInformation.IsOSPlatform(OSPlatform.Linux) ? PInvokeLinux.NvmlDeviceGetFanSpeed(device, ref fanSpeed)
: 12;
if (response != SUCCESS)
{
CUDA.Log(Level.Error, $"nvmlDeviceGetFanSpeed() responded with an error: {NvmlErrorString(response)}");
}
response =
RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? PInvokeWindows.NvmlDeviceGetTemperature(device, NvmlTemperatureSensors.Gpu, ref temperature)
: RuntimeInformation.IsOSPlatform(OSPlatform.Linux) ? PInvokeLinux.NvmlDeviceGetTemperature(device, NvmlTemperatureSensors.Gpu, ref temperature)
: 12;
if (response != SUCCESS)
{
CUDA.Log(Level.Error, $"nvmlDeviceGetTemperature() responded with an error: {NvmlErrorString(response)}");
}
response =
RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? PInvokeWindows.NvmlDeviceGetPowerUsage(device, ref power)
: RuntimeInformation.IsOSPlatform(OSPlatform.Linux) ? PInvokeLinux.NvmlDeviceGetPowerUsage(device, ref power)
: 12;
if (response != SUCCESS)
{
CUDA.Log(Level.Error, $"nvmlDeviceGetPowerUsage() responded with an error: {NvmlErrorString(response)}");
}
response =
RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? PInvokeWindows.NvmlDeviceGetClockInfo(device, NvmlClockType.Graphics, ref gpuClock)
: RuntimeInformation.IsOSPlatform(OSPlatform.Linux) ? PInvokeLinux.NvmlDeviceGetClockInfo(device, NvmlClockType.Graphics, ref gpuClock)
: 12;
if (response != SUCCESS)
{
CUDA.Log(Level.Error, $"nvmlDeviceGetClockInfo() responded with an error: {NvmlErrorString(response)}");
}
response =
RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? PInvokeWindows.NvmlDeviceGetClockInfo(device, NvmlClockType.Mem, ref vramClock)
: RuntimeInformation.IsOSPlatform(OSPlatform.Linux) ? PInvokeLinux.NvmlDeviceGetClockInfo(device, NvmlClockType.Mem, ref vramClock)
: 12;
if (response != SUCCESS)
{
CUDA.Log(Level.Error, $"nvmlDeviceGetClockInfo() responded with an error: {NvmlErrorString(response)}");
}
response =
RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? PInvokeWindows.NvmlDeviceGetUtilizationRates(device, ref utilization)
: RuntimeInformation.IsOSPlatform(OSPlatform.Linux) ? PInvokeLinux.NvmlDeviceGetUtilizationRates(device, ref utilization)
: 12;
if (response != SUCCESS)
{
CUDA.Log(Level.Error, $"nvmlDeviceGetUtilizationRates() responded with an error: {NvmlErrorString(response)}");
}
if (utilization.Gpu == 0u && utilization.Mem == 0u && fanSpeed == 0u && temperature == 0u && power == 0u && gpuClock == 0u && vramClock == 0u)
{
return(null);
}
return(new DeviceQuery()
{
FanSpeed = $"{fanSpeed} %",
Temperature = $"{temperature} C",
PowerDraw = $"{GetNormalizedPower(fullPciBusID, (decimal)power / 1000):F2} W",
ClockGPU = $"{gpuClock} MHz",
ClockVRAM = $"{vramClock} MHz",
UtilizationGPU = $"{utilization.Gpu} %",
UtilizationVRAM = $"{utilization.Mem} %"
});
}
