I’m trying to write a daemon for monitoring GPU power usage for a Tesla C2075. When using the NVML API, calls to nvmlDeviceGetPowerUsage always report 0 mW (with NVML_SUCCESS as the return value). Calls to other NVML API, such as nvmlDeviceGetTemperature, report non-zero numbers and also return NVML_SUCCESS. The nvidia-smi tool’s output is consistent with my own application’s behavior. What am I doing wrong with my C2075? Do power readings require special motherboard support? Here’s a brief overview of my setup (I’ve got the Tesla C2075 in a desktop-class motherboard):
Motherboard : ASUS M5A97
CPU : AMD FX-8350 8-core processor
GPU : NVIDIA Tesla C2075
OS : Ubuntu Server 14.04
Thanks for your help!