tensorflow-on-aws
Scripts to install TensorFlow on an AWS EC2 GPU Instance
These scripts are based on the excellent guide by @ramhiser available here
The first part takes about 10 mins and the second part takes about 30 mins to finish
Note about cuDNN
This script requires an Nvidia Accelerated Computing Developer Program account. If you do not have an account already, register for an account at: https://developer.nvidia.com Once accepted (typically takes about a day), download cuDNN from: https://developer.nvidia.com/rdp/assets/cudnn-65-linux-v2-asset Host this file on a private location and provide the URL when the script runs.
Note about versions
This script is for the following specific versions:
| Name | Version |
|---|---|
| TensorFlow | 0.9.0 |
| CUDA | 7.0 |
| cuDNN | 7.0 |
| Java | 8 |
| Bazel | 0.2.3 |
| Python | 2.7.6 |
Usage
- Request an AWS EC2 (spot) instance with the following settings:
- AMI: Ubuntu Server 14.04 LTS (HVM), SSD Volume Type - ami-fce3c696
- Type: g2.2xlarge
- Storage: 16 GB (or higher)
- SSH into the instance
- Download part 0 of 3 of the script
wget https://raw.githubusercontent.com/jasimpson/tensorflow-on-aws/master/toa_part_0of2.sh
- Set executable
chmod +x toa_part_0of2.sh
- Execute script part 0 of 2
./toa_part_0of2.sh
- Execute script part 1 of 2
./toa_part_1of2.sh
- Enter URL to download cuDNN from
- Reboot machine
sudo reboot
- SSH back into machine
- Execute script part 2 of 2
./toa_part_2of2.sh
- Test TensorFlow
$ python>>> import tensorflow as tf