That seems to be the case for GTX 1080 and CUDA toolkit 8.0. Below is the output when running Kaldi's cu-vector-speed-test:
Thu Apr 27 10:16:37 2017
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 375.39 Driver Version: 375.39 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 1080 Off | 0000:01:00.0 Off | N/A |
| 28% 33C P2 43W / 180W | 161MiB / 8114MiB | 31% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 25395 C ./cu-vector-speed-test 159MiB |
+-----------------------------------------------------------------------------+