X-ScaleAI is a high performance solution for distributed training for your complex AI problems:

Our Package

  • Exploiting HPC Technologies for Deep Learning
    on x86 and OpenPOWER Platforms
  • Fine-tuned CUDA-Aware MPI library for GPU clusters
  • Fine-tuned MPI library for CPU-only systems
  • Distributed Training with TensorFlow, Pytorch, and MXNet using Horovod
  • “Out of the box” optimal performance on OpenPOWER (POWER9) + GPU platforms, including #1 Summit system
  • “Out of the box” optimal performance on CPU-only platforms, including #5 Frontera system
  • Tested with ResNet-50 Training using TensorFlow benchmark on Frontera using up to 2,048 nodes and on Summit using up to 1,536 Volta GPUs
  • Simple installation and execution in one command
  • Coming soon: support for more system configurations

Our Performance

*Graph courtesy OSU-NBCL Tutorial

Interested in a free trial?