Intel CPU¶

When using Intel Xeon CPUs for training and inference, the mxnet-mkl package is recommended. Adding --pre installs a nightly build from master. Without it you will install the latest patched release of MXNet:

$ pip install mxnet-mkl [--pre]

Or build MXNet from source code with USE_MKLDNN=1. For Linux users, USE_MKLDNN=1 will be turned on by default.

We also find that setting the following environment variables can help:

Note that MXNet treats all CPUs on a single machine as a single device. So whether you specify cpu(0) or cpu(), MXNet will use all CPU cores on the machine.

Other CPU¶

If using CPUs (not just Intel CPUs – ARMs also), NNPACK can improve the running performance with 2x~7x, please check nnpack.md for details.

Nvidia GPU¶

cuDNN typically accelerates MXNet performance on NVIDIA GPUs significantly, especially for convolution layers. We suggest always checking to make sure that a recent cuDNN version is used.

Setting the environment export MXNET_CUDNN_AUTOTUNE_DEFAULT=1 sometimes also helps.

We show results when using various GPUs including K80 (EC2 p2.2xlarge), M60 (EC2 g3.4xlarge), and V100 (EC2 p3.2xlarge).

Multiple Devices¶

If more than one GPU or machine are used, MXNet uses kvstore to communicate data. It’s critical to use the proper type of kvstore to get the best performance. Refer to multi_device.md for more details.

Besides, we can use tools/bandwidth to find the communication cost per batch. Ideally, the communication cost should be less than the time to compute a batch. To reduce the communication cost, we can consider:

• Exploring different --kv-store options.
• Increasing the batch size to improve the computation to communication ratio.

Input Data¶

To make sure you’re handling input data in a reasonable way consider the following:

• Data format: If you are using the rec format, then everything should be fine.
• Decoding: By default, MXNet uses 4 CPU threads for decoding images. This is often sufficient to decode more than 1K images per second. If you are using a low-end CPU or your GPUs are very powerful, you can increase the number of threads.
• Storage location. Any local or distributed file system (HDFS, Amazon S3) should be fine. If multiple devices read the data from the shared network file system (NFS) at the same time, problems might occur.
• Use a large batch size. We often choose the largest one that fits into GPU memory. A value that’s too large can slow down convergence. For example, the safe batch size for CIFAR 10 is approximately 200, while for ImageNet 1K, the batch size can exceed 1K.

Profiler¶

As of v0.9.1 (with the NNVM merge), MXNet has a built-in profiler that gives detailed information about execution time at the symbol level. This feature complements general profiling tools like nvprof and gprof by summarizing at the operator level, instead of a function, kernel, or instruction level.

In order to be able to use the profiler, you must compile MXNet with the USE_PROFILER=1 flag in config.mk.

The profiler can then be turned on with an environment variable for an entire program run, or programmatically for just part of a run. See example/profiler for complete examples of how to use the profiler in code, but briefly, the Python code looks like:

    mx.profiler.set_config(profile_all=True, filename='profile_output.json')
    mx.profiler.set_state('run')

    # Code to be profiled goes here...

    mx.profiler.set_state('stop')

The mode parameter can be set to

• symbolic to only include symbolic operations
• all to include all operations

After the program finishes, navigate to your browser’s tracing (Example - chrome://tracing in a Chrome browser) and load the profile_output.json file output by the profiler to inspect the results.

Note that the output file can grow extremely large, so this approach is not recommended for general use.