Benchmark of opensource Platforms

Machine:

• Server: Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz, 2 Sockets, 20 Cores per socket

• CPU environment System: Ubuntu 16.04.3 LTS, Docker 17.05.0-ce, build 89658be

• GPU environment System: Ubuntu 16.04.3 LTS, NVIDIA-Docker 17.05.0-ce, build 89658be NVIDIA Docker image: nvidia/cuda:8.0-cudnn5-devel-ubuntu16.04

PaddlePaddle: 0.11.0(Fluid)

• paddlepaddle/paddle:latest

TensorFlow: 1.4.0

• tensorflow/tensorflow:latest

Benchmark Model

selected models PaddlePaddle Fluid vs TensorFlow

We selected some classic models, compare the performance and speed with TensorFlow.

	train cost	train accuracy	test accuracy	samples/sec	train cost	train accuracy	test accuracy	samples/sec
MNIST CNN
VGG-19
RESNET-101
Stacked LSTM

• TBD add charts compare here

• VGG-19 input image size - 3 * 224 * 224, Time: images/second

BatchSize	64	128	256
PaddlePaddle Fluid
TensorFlow

• TBD add charts compare here

• RESNET-101

BatchSize	64	128	256
PaddlePaddle Fluid
TensorFlow

• TBD

add charts here

• Stacked LSTM

BatchSize	64	128	256
PaddlePaddle Fluid
TensorFlow

• TBD

add charts here

PaddlePaddle books Fluid vs Paddle 0.10.0

To validate the Fluid performance on general models, we choose the models in book chapter, compare the performance and speed with Paddle 0.10.0.

	train cost	train accuracy	test accuracy	samples/sec	train cost	train accuracy	test accuracy	samples/sec
01.fit_a_line
02.recognize_digits
03.image_classification
04.word2vec
05.recommender_system
06.understand_sentiment
07.label_semantic_roles
08.machine_translation