Skip to main content

CPU-Only Instance Types

· 2 min read

Save even more money with CPU-Only job instances.

How It Works​

CPU-Only instances can be provisioned in increments of 4 vCPU. Each increment of vCPU has a minimum of 10GB of RAM available. The same built-in trainML job environments can be used for CPU-Only jobs, but have some GPU-specific libraries disabled.

Using the Web Platform​

To create a CPU-only job, select None (CPU-Only) as the GPU Type in the job form. You will then have the ability to specify the CPU Count field instead of the GPU Count field. Click Next to review your job configuration and Create to create the job.

info

If a Notebook is created as a CPU-Only notebook, a GPU cannot be attached to it later through the Edit functionality. Likewise, a Notebook created with a GPU attached cannot be converted to a CPU-Only notebook. However, you can use the Copy functionality to create a new notebook and attach/detach a GPU at that time.

Using the SDK​

To create a CPU-Only job with the SDK, specify cpu as the only entry in the gpu_types property.

job = await trainml.jobs.create(
    "CPU Training Job",
    type="training",
    gpu_types=["cpu"],
    cpu_count=8,
    disk_size=10,
    ...
)
info

The cpu GPU Type cannot be intermingled with other GPU types. Additionally, if the GPU type is set to cpu, the gpu_count field cannot be provided.