Open sandeep-krishnamurthy opened 6 years ago
I would also suggest to extend the C implementation to support wider GPU info. More precisely I suggest to add number of GPUs available into output of this method. And then, wrap it all up and expose via Python API.
Platform independent API (without using nvidia-smi) for querying GPU memory. C implementation is provided by @sbodenstein in this PR - https://github.com/apache/incubator-mxnet/pull/12083#pullrequestreview-145347358