ml-energy / zeus

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

Test and verify `nvmlDeviceSetAPIRestriction` #59

Open jaywonchung opened 2 months ago

jaywonchung commented 2 months ago

nvmlDeviceSetAPIRestriction seems to offer a way to reduce the permissions required by setting the GPU's frequency (but not power limit). If this works, a workflow could be for a administrator to (1) turn on GPU persistence mode and (2) reduce restrictions for application clock setting, and Perseus would in theory work with no extra privileges. If this works fine, I should consider implementing GlobalFrequencyOptimizer, which is the same as GlobalPowerLimitOptimizer but with GPU frequency.