About Choosing a Compute Shape for Machine Learning

If you want to choose the compute shape that's best suited to your anticipated machine learning workloads, you can conduct an iterative process that identifies the optimum shape.

A shape is a template that determines the number of CPUs, number of GPUs, amount of memory, and other resources that are allocated to an instance.

The available shapes cover various cost and performance options, and choosing a shape that best suits your needs can be challenging. The task is made easier and faster by first applying a few rules of thumb and then conducting a systematic search to select the shape that best suits your needs.

One of the terms that you'll come across when choosing a shape is OCPU. Oracle defines an OCPU as the CPU capacity equivalent of one physical core of an Intel Xeon processor with hyper threading enabled. For Intel Xeon processors, each OCPU corresponds to two hardware execution threads, known as vCPUs.

The GPU shapes that are available on Oracle Cloud Infrastructure have NVIDIA Pascal or NVIDIA Volta-based GPUs, in bare metal and in virtual machine configurations.

Prerequisites

Before you can start selecting an appropriate shape, you must first have access to the available GPU shapes. You should also have one or more representative machine learning workloads ready.

GPU shapes might not be initially available due to service limits, which are usually established with your Oracle sales representative when you purchase Oracle Cloud Infrastructure. You can request to have the service limits raised to include GPU instances. The Oracle Cloud Infrastructure documentation contains instructions on how to do that.

You should also have some machine learning workloads ready. The workloads can be benchmark tests, but preferably they should be machine learning tasks that use the data and methods that you plan to employ on your selected compute instance. If your dataset is very large and you want to speed up the process of selecting a shape, you can use a representative subset for the tests and extrapolate the results.

About Required Services and Products

You must have access to Oracle Cloud Infrastructure Compute with GPU shapes available.

If you don't already have access to Oracle Cloud Infrastructure Compute, see Learn how to get Oracle Cloud services for Oracle Solutions.

If no GPU shapes are available in your tenancy, then you must request to have your service limit raised.

To request a service limit increase:

1. Log into Oracle Cloud console.
2. Open the Help menu and click Request service limit increase.
3. Complete the form.

Most service limit requests are resolved within one business day. However, depending on the resource or quantity requested, it might take longer to allocate.