When you train or score a model or function, you choose the type, size, and power of the hardware configuration that matches your computing needs.
Choose the hardware configuration for your Watson Machine Learning asset when you train the asset or when you deploy it.
| Capacity type | Capacity units per hour |
|---|---|
| Extra small: 1x4 = 1 vCPU and 4 GB RAM | 0.5 |
| Small: 2x8 = 2 vCPU and 8 GB RAM | 1 |
| Medium: 4x16 = 4 vCPU and 16 GB RAM | 2 |
| Large: 8x32 = 8 vCPU and 32 GB RAM | 4 |
| Extra large: 16x64 = 16 vCPU and 64 GB RAM | 8 |
Deployments and scoring consume compute resources as capacity unit hours (CUH) from the Watson Machine Learning service.
You can monitor the total monthly amount of CUH consumption for the Watson Machine Learning service on the Environments page.
The rate of capacity units per hour consumed is determined by the computing requirements of your deployments, based on such variables as type of deployment and framework as well as the complexity of scoring. Note that scaling a deployment to support more concurrent users and requests also increases CUH consumption. Because there are so many variables that affect resource consumption for a deployment, we recommend you run tests on your models and deployments to analyze CUH consumption.
The way that online deployments consume capacity units is based on framework. For some frameworks, CUH is charged for the number of hours the deployment asset is active in a deployment space. For example, SPSS models in online deployment mode that run 24 hours a day for seven days a week consume CUH and are charged for that period. There is no idle time for an active online deployment. For other frameworks, CUH is charge according to scoring duration. See the CUH consumption table for details on how CUH is calculated.
Compute time is calculated to the millisecond. However, there is a one-minute minimum for each distinct operation. That is, a training run that takes 12 seconds is billed as one minute toward the capacity unit hour quota, while a training run that takes 83.555 seconds is billed exactly as calculated.
CUH consumption is calculated using these formulas:
| Deployment type | Framework | CUH calculation |
|---|---|---|
| Online | AutoAI, AI function, SPSS, Scikit-Learn custom libraries, Tensorflow, RShiny | deployment_active_duration no_of_nodes CUH_rate_for_capacity_type_framework |
| Online | Spark, PMML, Scikit-Learn, Pytorch, XGBoost | score_duration_in_seconds no_of_nodes CUH_rate_for_capacity_type_framework |
| Batch | all frameworks | job_duration_in_seconds no_of_nodes CUH_rate_for_capacity_type_framework |