Experience the benefits of having access to a hybrid cloud solution. Using Cloudera Machine Learning (CML), on the Cloudera Data Platform (CDP), see how an AI workload compares running on-premises versus leveraging computational resources in the cloud.
There are two (2) options in getting assets for this tutorial:
It contains only necessary files used in this tutorial. Unzip tutorial-files.zip and remember its location.
It provides assets used in this and other tutorials; organized by tutorial title.
In the ML Workspaces section, select Provision Workspace.
Two simple pieces of information are needed to provision an ML workspace - the Workspace Name and the Environment name. For example:
Beginning from the ML Workspaces section, open your workspace by selecting its name,
Select New Project.
Complete the New Project form using:
A project showcasing the speed improvements of running heavy AI workloads on-premises versus using GPU resources on the cloud.
Select Create Project
We will create three (3) experiments to verify speed improvements of AI workload and see the effect GPUs have on training the model.
Beginning from the Projects section, select the project name,
In the Experiments section, select Run Experiment and complete the form as follows:
Similarly, let’s create an experiment using 1 GPUs:
Similarly, let’s create an experiment using 2 GPUs:
As the experiment results were completing, you could see an order of magnitude difference between having access to GPUs and having to train the model on CPU only.
Your results should be similar to:
The training time utilized for 0 GPU should be comparable to on-premises with no GPUs.
You can review the output of the python program, main.py, by selecting a Run id, then select Session.
Congratulations on completing the tutorial.
As you’ve now experienced, having access to a hybrid cloud solution allows the opportunity to leverage cloud resources only when you need them. In our experiments, the use of GPUs resulted in huge time savings, empowering users to spend their valuable time creating value instead of waiting for their model to train.
This may have been caused by one of the following: