nvidia-smi freeze

There was a problem with the call nvidia-smi.
To monitor the load on the GPU M60 I use nvidia-smi and read the required values.

The GPU is used to transcode video using ffmpeg.

The command of the nvidia-smi is performed every second, and with a sufficiently low load on the GPU ~ 15%, the response time of the request increases to 4.236s
In this case, requests are not blocked and accumulates a lot of processes that in turn cause problems with the GPU.

Team example:

nvidia-smi -q -d UTILIZATION, MEMORY
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 390.25                 Driver Version: 390.25                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Tesla M60           On   | 00000000:06:00.0 Off |                  Off |
| 32%   49C    P0    46W / 120W |    977MiB /  8129MiB |      1%      Default |
+-------------------------------+----------------------+----------------------+
|   1  Tesla M60           On   | 00000000:07:00.0 Off |                  Off |
| 32%   38C    P0    46W / 120W |    814MiB /  8129MiB |      1%      Default |
+-------------------------------+----------------------+----------------------+
|   2  Tesla M60           On   | 00000000:86:00.0 Off |                  Off |
| 35%   52C    P0    46W / 120W |    840MiB /  8129MiB |      1%      Default |
+-------------------------------+----------------------+----------------------+
|   3  Tesla M60           On   | 00000000:87:00.0 Off |                  Off |
| 32%   43C    P0    51W / 120W |   1336MiB /  8129MiB |      1%      Default |
+-------------------------------+----------------------+----------------------+

How often can I request a load data, or how can I optimize the command for faster interaction.

Try using the persistence daemon to keep the gpus initialized.

After each OS restart, I initialize the command

nvidia-persistenced
nvidia-smi --auto-boost-default=0

But still at a load, the call time of the team is greatly increased.