Benchmark says that the GPU has only 1.7GB memory

Hello guys,

I need some help. I’m currently doing AI-Benchmark. this is a benchmark that checks 19 diffrent models on inference or/and training. I need 2GB RAM for inference and 4GB for training so training won’t be happening but inference should be able to complete except it doens’t it says the GPU has only 1.7GB RAM while it should be shared memory

image

At some point the benchmark crashes due to having not enough memory.

do you guys have any idea how to fix this?

Some extra info:

Hi,

Since Jetson’s memory is shared between CPU and GPU, it’s possible that the GPU has available memory like yours.
To confirm this, could you create a TensorFlow session and monitor the memory status with tegrastats?

$ sudo tegrastats

Thanks.