Hello,
I am trying to run the nnet3 setting on my laptop with a single GPU (GeForce GTX 960M).
I guess it is not the wisest option to do so as it will take a lot of time, but anyway I am struggling to make it work.
I have installed CUDA 9.2 and nvidia-smi returns:
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 396.26 Driver Version: 396.26 |
|-------------------------------+----------------------+----------------------+
| 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 960M Off | 00000000:01:00.0 Off | N/A |
| N/A 44C P0 N/A / N/A | 0MiB / 4046MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
I have also compiled Kaldi src with CUDA enabled and it didn't throw any errors.
For some reason, while running steps/nnet3/train_dnn.py I keep getting an error:
ERROR ([5.3.81~1-829b0]:SelectGpuId():cu-device.cc:121) No CUDA GPU detected!, diagnostics: cudaError_t 30 : "unknown error", in cu-device.cc:121
I found out that the error is also thrown when I simply run the nnetbin/cuda-gpu-available... What can be the cause of this issue?
Best regards,
Artur Zygadło