A few weeks ago we published a tutorial on Easy TensorFlow Model Training on AWS using our Bitfusion TensorFlow AMI. This quick tutorial as well as the AMI have proven immensely popular with our users and we received various feature requests. As such this week we are releasing v0.03 of the TensorFlow AMI which introduces several new features:
- TensorFlow 0.9
- cuDNN 5
- 30-40% performance improvement over previous AMI
- Keras Deep Learning Library
- Github based documentation for easier reference
- Python 3 support
It is always good to do some regression testing when adding new features and so once again we decided to run the multi-gpu Cifar10 model training example which we demonstrated in the previous tutorial. We utilized the g2.8xlarge instance on AWS and benchmarked the performance across one to four GPUs. The graph below shows the results comparing our previous TensorFlow 0.8 AMI with cuDNN 4 vs. our new TensorFlow 0.9 AMI with cuDNN 5. Remarkably, on similar hardware the new AMI performs between 30% to 40% faster than the previous release.
If you’ve been stumped by long run-times when training your models, this alone may be a good incentive to upgrade to the new AMI release.
Questions or comments? Please post them in the comment section below or join our community Bitfusion-AWS Slack Channel.