Force P2 for CUDA workload

I have an application running on some 2080Ti GPUs. While running I observe that sometimes the GPUs go into P8 power state while there is some idle time, then switch back to P2 when they are used again. This causes some latency (~20ms) from switching power states that I’d like to avoid. Is there some way to force the GPUs to always stay in P2 and never go to P8?

I found another thread about this: Avoiding performance mode slowdown
The solution there was to keep a kernel running continuously (say every 1s) to keep the GPU from going to P8. That works, but is a stupid solution, so I’d like to avoid that. There should be some other way to keep the GPUs from going to the lower power state.