How we score your device


Orion provides a benchmark scoring system to assess the deep learning capability of computing devices. Current scoring produces the weighted performance of individual server by executing a few classical machine-learning tasks under Orion virtual environment(same as task-executing environment), more specifically, in terms of image classification. Results are showed as the number of images the device will process per second. The benchmark explains how fast a given device(CPU+GPU) will perform standard deep learning tasks, and will be a point of reference for cross-platform device comparison. The more images/sec a device generates during benchmark testing, the higher performance it gives. For example, a dual 1070Ti GPU server outperforms single 1080Ti in terms of benchmark result, and may be more preferably chosen by developers. In order to satisfy compute requirements, all devices will be asked to perform benchmark test at the first time of client configuration. Together with other system parameters(memory, I/O speed, disk storage) , device scoring allows workers to be grouped to each category.


  • CPU: 6-core Intel-8700k @ 4.7GHz
  • GPU: Nvidia GTX Geforce Series
  • OS: Ubuntu 16.04 LTS with testing run via Docker version nbai-tensorflow1.5
  • Tensorflow/CUDA/cuDNN: 1.5.0/9.0/5.15
  • Dataset: ImageNet
  • Disk: Local SSD 256 GB

Sample Score and Price

Orion GPU Instance
Benchmark Scores