I’m trying to lock the memory clock of my GPUs (2080ti, 2060 super) to do some benchmarking, after a few days of googling around and searching on NVIDIA docs, I still can’t figure out how to do that.
What I have tried:
- using
nvmlDeviceSetApplicationsClocks
, but it shows NVML_ERROR_NOT_SUPPORTED - using
nvidia-smi lmc
, also shows NVML_ERROR_NOT_SUPPORTED - trying to find a way to force performance state (still can’t do that for now.)
(note: those APIs work on Ampere cards)
There are plenty of crypto miners and crypto OS (with linux based like mmpos.eu/hiveos) in the wild that are able to lock the memory clock on Pascal/Turing cards at a few points (810Mhz, 3500Mhz, 5001Mhz,…). Sadly, they are all closed source.
Is there any official way to do that via nvml in C/C++?