Quota policy

The IBM Watson Machine Learning engine enforces quotas on a per-instance basis to ensure appropriate resource allocation and usage. These policies vary according to account profiles, historical service usage, and availability of resources. The quota policy describes limits that are not defined by the pricing plan and are subject to change without notice. The following service quota limits are currently active.

Resource Usage

Resources are limited to the following maximum values:

  • Max published models: 200 (Lite Plan) 1000 (Standard & Professional Plans)
  • Max Spark ML model size: 200 MB
  • Max deployed models: 1000 (Standard & Professional Plans)

Note: To understand Lite Plan limits, see the description of the Lite Plan in the IBM Cloud Catalog.

Service Request Limits

You are limited in the number of requests that you can make per minute to specific APIs. If your application needs higher limits, you must contact IBM Cloud Support to increase your quota.

Deployment requests

The following limits apply to deployment API requests:

  • Period: 60 seconds
  • Default limit: 10

Online prediction requests

The following limits apply to online predictions requests:

  • Period: 60 seconds
  • Default limit: 500

Resource management requests

The following limits apply to the combined total requests in this list:

Token: post, get, delete, put, patch requests

Deployments: post, get, delete, put, patch requests

  • Period: 60 seconds
  • Default limit: 100

Resource Usage for Deep Learning Training

Resources for Deep Learning Training can be allocated in the following way:

  • Max number of GPUs: 8 (Lite Plan)
  • Max number of GPUs for distributed training: 8 (Standard & Professional Plan)
  • Type of GPUs: k80 (Lite Plan)
  • Type of GPUs for distributed training: p100 (Standard & Professional Plan)

For more information about deep learning, see Deep Learning Overview

Learn more

For more information about the API, see REST API.

For more information about IBM® SPSS® Modeler and the modeling algorithms it provides, see IBM Knowledge Center.