ml-energy / zeus

Deep Learning Energy Measurement and Optimization
https://ml.energy/zeus
Apache License 2.0
179 stars 24 forks source link

Distinguish instantaneous power vs. average power #45

Closed jaywonchung closed 1 month ago

jaywonchung commented 3 months ago

For ZeusMonitor's approx_instant_energy, we need to read the instantaneous power of the GPU.

It seems like NVML decided to default to average power for GPUs newer than GA100. (source) We need to look into nvmlDeviceGetFieldValues with field value designated as instant power. Double check whether this is backwards compatible.