I have GTX 1050 mobile graphic card. I have set up my tensorflow-gpu with conda, and have managed to get my tensorflow code work using GPU.
In the Windows Task Manager, I found my GPU has two compute engine metric, compute_0 and compute_1. But when I try to train my NN, I found only the compute_0 took the workload, and the other one was always at rest.
This makes me wonder if:
- My system is set up correctly.
My code is set up correctly.
Honestly, I am not completely sure if compute_0 and compute_1 refer to the CUDA core computation workload. Any suggestion about CUDA set up checking and hardware verification are desired.