Why did my rcnn function not recognize the GPU device on Amazon EC2 P2 instance I connected to via the cloud center?

I followed the step-by-step documentation (Deep learning with MATLAB and multiple GPUs) to connect to Amazon EC2-P2 instance and I have tested and validated the cloud connection. Everything seemed fine. However, when I tried to run trainRCNNOBjectDetector, I got an error - "GPU support for convolutional neural networks require a GPU device with compute capability 3.0 or higher." Could anyone suggest how to debug and solve this issue?

[Walter Roberson]

(I just looked up and found that P2 are advertised as having K80 GPU, and I find that Tesla K80 have Compute Capacity 3.7. Which should be fine if only they could be recognized.)

[Joss Knight]

I suspect you just haven't configured your environment to run on the cluster, so it is trying to run on the client. Have you set the cluster as your default cluster? Have you started a pool on that cluster using parpool? Have you enabled UseParallel in the Computer Vision System Toolbox preferences?

https://www.mathworks.com/help/vision/gs/computer-vision-system-toolbox-preferences.html