The settings are controlled by setting applications clocks (-ac) and optionally setting power limits (-pl) using the nvidia-smi tool. Please use the command-line help for nvidia-smi to get started.
By reducing application clocks, you can force the V100 to run more slowly than the maximum possible speed, which will push the GPU into its most efficient operating range (e.g. flops/watt)
There aren’t predefined “max P” (max performance) and “max Q” (max efficiency) settings provided, however, and I wouldn’t be able to tell you much about how to discover these. It should be possible to design a simple test, eg. run a benchmark such as matrix multiply and then vary the application clocks, to determine max-Q. Max P mode is the mode the GPU should be in by default.