Nimbix is pleased to announce the availability of TensorFlow™ with GPUs in the Nimbix Cloud powered by Jupyter Notebook in our application library. You can get started with TensorFlow™ in under a minute and run GPU-powered Jupyter Notebooks securely in your browser with no user installation necessary. Google has also released a free course for Deep Learning with TensorFlow™ on Udacity. The course is structured around interactive learning exercises with Jupyter Notebook which are made accessible in every Nimbix User’s environment upon starting a TensorFlow™ Jupyter Notebook. No user installation is required!
TensorFlow™ is a highly efficient library for numerical analysis in Python. Under the hood, it has an efficient computational back end implemented in C++. This, combined with an efficient data representation, has made TensorFlow™ one of the most powerful and easy to use numerical analysis libraries for Deep Learning. It also has great GPU and CPU compatibility for high performance computing environments. On the Nimbix Cloud, your Jupyter Notebook environment is configured to use the CPU or GPU version of TensorFlow 0.8 depending on the hardware you select in the task builder with no additional installation by the user. You can change between hardware configurations as needed without have to install anything. Simply launch the application from the application library under the Compute tab. To make it as easy as possible to get started with TensorFlow™, we have released a Jupyter Notebook-powered application, which provides a fully functional development environment directly in your browser.
Create a Nimbix Cloud Account
If you don’t already have an account on https://platform.jarvice.com, head over to to the Sign Up Page. You will need to enter your billing information prior to running a job using our Pay-As-You-Go service. We also have enterprise subscriptions available for production use cases. See our Pricing Page for more details.
Launch Jupyter Notebooks for TensorFlow™
Head over to the Compute tab once you have logged in to your account. Scroll down to click on TensorFlow™ and select the TensorFlow™ Jupyter Notebook. This will take you to the Task Builder screen which is the main entry point for configuring your hardware. Launching Jupyter Notebook provides you a fully-enabled Jupyter Notebook environment. TensorFlow is distributed with CPU and GPU versions. JARVICE, the software that powers the Nimbix Cloud, launches Jupyter Notebook with the correct version of TensorFlow for your hardware selection. TensorFlow is compatible with Python 2 and Python 3, so select the version of Python you would like your notebooks to use.
Configure your hardware by selecting from the available machine types. We currently have NVIDIA K40 and NVIDIA K80 GPUs available, as well as a Bitfusion Boost-enabled machine type. TensorFlow is also compatible with CPU machine types starting from $0.36 per hour. Press the Continue button and Submit to launch your job.n
Connecting to Jupyter Notebook
Your job will launch in a matter of seconds. Head over to the Dashboard to view your currently running jobs. Click on the job to expand the details, and when your notebook is ready, a green “Connect” button will appear. Before clicking connect, copy the password so you can paste it into the authentication window.
This will take you to the authentication screen, where you can enter the username: nimbix and password from your Dashboard. This will take you to the Jupyter Notebook interface directory in your browsers. The files displayed in Jupyter notebook are part of your user’s persistent storage vault which will be accessible from any job you run.
The first time you run TensorFlow Jupyter Notebook, the tensor-flow examples will be copied to your storage vault.
Udacity Course Files
The Udacity Deep Learning IPython Notebook files are availalble in tensorflow-examples/udacity. Try loading the first notebook, 01_notmnist.ipynb to get started with TensorFlow. All of the Udacity course’s python dependencies are already installed, so you are ready to go with TensorFlow without any additional installation.
Custom IPython Files
You can create your own IPython project files from the New tab in the top right corner of the Jupyter Project file browser and get started hacking away. Jupyter even has tab completion for the coding environment, which makes learning TensorFlow even easier.
Terminating your Jobs
When you are done running your work inside of the Jupyter Notebook environment, you need to terminate the job from your job dashboard. Click the “shutdown” button on the dashboard. Once the Processing status transitions to Completed, billing for your machine type stops with to-the-minute granularity. Wasn’t that easy?
Uploading and Downloading Custom Data
If you would like to upload or download, you can use rsync, scp or ssh using the Connection Address and password in your job’s details using the username nimbix. You can also access your data while your jobs are not running by using sftp. See our related documentation for uploading and downloading data.
TensorFlow™ Batch Jobs
We have also configured TensorFlow for running Python Scripts in batch. When you upload your data to your drop.jarvice.com vault, you will see your script in the file drop down on the TaskBuilder.
- Official TensorFlow Documentation
- Jupyter Project Documentation (No installation required on JARVICE)
- Google TensorFlow Deep Learning Course on Udacity
- JARVICE 2.0 Quick Start Guide (for developers of custom images)
- Transferring Data to and from JARVICE
Feel free to reach out to us with comments at firstname.lastname@example.org or Tweet us @Nimbix. We are always excited to learn more from you about how you integrate GPU and HPC jobs into your production pipelines.