Dataset Used

Dataset Used

Configuration

bash pre-install.sh

End-to-end benchmarking

For end-to-end benchmarking, a series of deep learning frameworks are required. Please refer to the Frameworks for framework information and installation.

Micro Benchmark

The micro benchmark is based on the DeepBench (By Baidu Research), primarily covering the deep learning components, such as Dense Matrix Multiplies, Convolutions, Recurrent Layers, and All Reduce. Please refer to the microbench folder for more information.

Workloads

MNIST

CIFAR-10

CIFAR-100

System Configuration:

• Operating System: Ubuntu 16.04
• Memory: 10 GB
• GPU: Nvidia GTX 1080 Ti (11GB)
• CPU: E5-1620 (3.6GHz)
• Hard Drive: 256 GB