Is nvmlDeviceGetPowerUsage() working with CUDA 4.0 and C2050? Do all the NVML APIs work with the lat

Accelerated Computing CUDA CUDA Programming and Performance
1

Is anyone using the NVML APIs to get power usage on a Tesla C2050? Most NVML functions work just fine, but the following returns NVML_ERROR_NOT_SUPPORTED:

unsigned int milliWattsInUse = 0 ;

rc = nvmlDeviceGetPowerUsage( deviceHandle, &milliWattsInUse );

if ( NVML_SUCCESS != rc )
{
// call fails with NVML_ERROR_NOT_SUPPORTED
}
else
{
// success is never returned
}

Even the ‘nvidia-smi -q’ output shows the power information as ‘N/A’ for both GPUs:

Power Readings
    Power State             : P12
    Power Management        : N/A
    Power Draw              : N/A
    Power Limit             : N/A

The sysadmin hasn’t updated to the final production driver yet (we’ll test this again when he does), but nvmlDeviceGetPowerUsage() and nvmlDeviceGetPowerCappingMode() seem to be disabled in toolkit 4.0RC2.

Other info: gcc 4.4.1, SUSE 11, Tesla C2050 boards, driver version 270.41.06

If anyone has any insight or can test this on the production release of toolkit 4.0, it would be greatly appreciated!

Thanks in advance,

td

2

Any updates?

Can you please post the complete source code?

3

To get the power readings power management mode needs to be enabled which is not supported in s2050 so unless you have a s2075 which actually has power management enabled you cannot get power readings using either nvml library or nvidia-smi