Some Tips for Improving MXNet Performance

Even fixing the training or deployment environment and parallelization scheme, a number of configuration settings and data-handling choices can impact performance. In this document, we address some tips for improving MXNet performance.

Performance is mainly affected by the following 4 factors:

1. Implementation of operators (Convolution, Pooling, ..)
   • Intel CPU
   • Nvidia GPU
2. Input data loading and augmentation
   • Input Data
3. Workloads (computation graph) optimization and scheduling
   • Profiler
4. Communication for multi-devices training
   • Multiple Devices

Intel CPU

For using Intel Xeon CPUs for training and inference, we suggest enabling both USE_MKL2017 = 1 and USE_MKL2017_EXPERIMENTAL = 1 in config.mk. Check MKL_README.md for details.

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

• export KMP_AFFINITY=granularity=fine,compact,1,0 if there are two physical CPUs
• export OMP_NUM_THREADS=vCPUs / 2 in which vCPUs is the number of virtual CPUs. Whe using Linux, we can access this information by running cat /proc/cpuinfo | grep processor | wc -l

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.

Scoring results

The following table shows performance, namely number of images that can be predicted per second. We used example/image-classification/benchmark_score.py to measure the performance on different AWS EC2 machines.

AWS EC2 C4.8xlarge:

Batch	Alexnet	VGG	Inception-BN	Inception-v3	Resnet 50	Resnet 152
1	119.57	34.23	111.36	54.42	42.83	19.51
2	210.58	51.63	137.10	67.30	57.54	23.56
4	318.54	70.00	187.21	76.53	63.64	25.80
8	389.34	77.39	211.90	84.26	63.89	28.11
16	489.12	85.26	220.52	82.00	63.93	27.08
32	564.04	87.15	208.21	83.05	62.19	25.76

AWS EC2 C4.4xlarge:

Batch	Alexnet	VGG	Inception-BN	Inception-v3	Resnet 50	Resnet 152
1	109.96	23.00	71.82	28.10	30.66	11.81
2	124.56	24.86	81.61	31.32	32.73	12.82
4	157.01	26.60	86.77	32.94	33.32	13.16
8	178.40	30.67	88.58	33.52	33.32	13.32
16	189.52	35.61	90.36	33.63	32.94	13.18
32	196.61	38.98	105.27	33.77	32.65	13.00

AWS EC2 C4.2xlarge:

Batch	Alexnet	VGG	Inception-BN	Inception-v3	Resnet 50	Resnet 152
1	70.75	12.87	42.86	16.53	18.14	7.01
2	71.53	13.08	45.66	17.38	18.53	7.18
4	84.72	15.38	47.50	17.80	18.96	7.35
8	93.44	18.33	48.08	17.93	18.99	7.40
16	97.03	20.12	55.73	18.00	18.91	7.36
32	113.90	21.10	62.54	17.98	18.80	7.33

AWS EC2 C4.xlarge:

Batch	Alexnet	VGG	Inception-BN	Inception-v3	Resnet 50	Resnet 152
1	37.92	6.57	23.09	8.79	9.65	3.73
2	36.77	7.31	24.00	9.00	9.84	3.78
4	43.18	8.94	24.42	9.12	9.91	3.83
8	47.05	10.01	28.32	9.13	9.88	3.83
16	55.74	10.61	31.96	9.14	9.86	3.80
32	65.05	10.91	33.86	9.34	10.31	3.86

AWS EC2 C4.large:

Batch	Alexnet	VGG	Inception-BN	Inception-v3	Resnet 50	Resnet 152
1	19.86	3.67	12.20	4.59	5.11	1.97
2	19.37	4.24	12.41	4.64	5.15	1.98
4	22.64	4.89	14.34	4.66	5.16	2.00
8	27.19	5.25	16.17	4.66	5.16	1.99
16	31.82	5.46	17.24	4.76	5.35	OOM
32	34.67	5.55	17.64	4.88	OOM	OOM

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), M40, and P100 (DGX-1).

Scoring results

Based on example/image-classification/benchmark_score.py and MXNet commit 0a03417, with CUDNN 5.1

• K80 (single GPU)

  Batch	Alexnet	VGG	Inception-BN	Inception-v3	Resnet 50	Resnet 152
  1	202.66	70.76	74.91	42.61	70.94	24.87
  2	233.76	63.53	119.60	60.09	92.28	34.23
  4	367.91	78.16	164.41	72.30	116.68	44.76
  8	624.14	119.06	195.24	79.62	129.37	50.96
  16	1071.19	195.83	256.06	99.38	160.40	66.51
  32	1443.90	228.96	287.93	106.43	167.12	69.73

• M40

  Batch	Alexnet	VGG	Inception-BN	Inception-v3	Resnet 50	Resnet 152
  1	412.09	142.10	115.89	64.40	126.90	46.15
  2	743.49	212.21	205.31	108.06	202.17	75.05
  4	1155.43	280.92	335.69	161.59	266.53	106.83
  8	1606.87	332.76	491.12	224.22	317.20	128.67
  16	2070.97	400.10	618.25	251.87	335.62	134.60
  32	2694.91	466.95	624.27	258.59	373.35	152.71

• P100

  Batch	Alexnet	VGG	Inception-BN	Inception-v3	Resnet 50	Resnet 152
  1	624.84	294.6	139.82	80.17	162.27	58.99
  2	1226.85	282.3	267.41	142.63	278.02	102.95
  4	1934.97	399.3	463.38	225.56	423.63	168.91
  8	2900.54	522.9	709.30	319.52	529.34	210.10
  16	4063.70	755.3	949.22	444.65	647.43	270.07
  32	4883.77	854.4	1197.74	493.72	713.17	294.17

Training results

Based on example/image-classification/train_imagenet.py and MXNet commit 0a03417, with CUDNN 5.1. The benchmark script is available at here, where the batch size for Alexnet is increased by 8x.

• K80 (single GPU)

  Batch	Alexnet(*8)	Inception-v3	Resnet 50
  1	230.69	9.81	13.83
  2	348.10	15.31	21.85
  4	457.28	20.48	29.58
  8	533.51	24.47	36.83
  16	582.36	28.46	43.60
  32	483.37	29.62	45.52

• M40

  Batch	Alexnet(*8)	Inception-v3	Resnet 50
  1	405.17	14.35	21.56
  2	606.32	23.96	36.48
  4	792.66	37.38	52.96
  8	1016.51	52.69	70.21
  16	1105.18	62.35	83.13
  32	1046.23	68.87	90.74

• P100

  Batch	Alexnet(*8)	Inception-v3	Resnet 50
  1	809.94	15.14	27.20
  2	1202.93	30.34	49.55
  4	1631.37	50.59	78.31
  8	1882.74	77.75	122.45
  16	2012.04	111.11	156.79
  32	1869.69	129.98	181.53

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 mutli_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. Once enabled, the profiler can be enabled 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.profiler_set_config(mode='all', filename='profile_output.json')
    mx.profiler.profiler_set_state('run')

    # Code to be profiled goes here...

    mx.profiler.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 chrome://tracing in a Chrome browser and load the .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.