Managing hardware specifications for deployments

When you deploy certain assets in Watson Machine Learning, you can choose the type, size, and power of the hardware configuration that matches your computing needs.

Creating hardware specifications for deployments

You can create hardware specifications for your deployments in the following ways:

  • Python client library: Use the hardware_specifications.store function from the Python client library. For more information, see Python client library reference
  • Data and AI Common Core API: Use POST /v2/hardware_specifications from the Environments list in the Data and AI Common Core API to create a hardware specification. For more information, see Environments API reference.
  • cpd-cli environment hardware: Use the cpd-cli environment hardware-specification create function to create a hardware specification from cpd-cli. For more information, see cpd-cli command reference.

Deployment types that require hardware specifications

Selecting a hardware specification is available for all batch deployment types. For online deployments, you can select a specific hardware specification if you're deploying:

  • Python Functions
  • Tensorflow models
  • Models with custom software specifications

Hardware configurations available for deploying assets

  • XXS: 1x2 = 1 CPU and 2 GB RAM
  • XS: 1x4 = 1 CPU and 4 GB RAM
  • S: 2x8 = 2 CPU and 8 GB RAM
  • M: 4x16 = 4 CPU and 16 GB RAM
  • L: 8x32 = 8 CPU and 32 GB RAM
  • XL: 16x64 = 16 CPU and 64 GB RAM

You can use the XXS and XS configurations to deploy:

  • Python functions
  • Python scripts
  • R scripts
  • Shiny apps
  • Models based on custom libraries and custom images

For Decision Optimization deployments, you can use these hardware specifications:

  • S
  • M
  • L
  • XL

Hardware specifications for GPU inferencing

Beginning Cloud Pak for Data version 4.8.5, you can select GPU hardware specifications for CUDA software specifications from the user interface on x86 platform when you create a deployment. For more information, see Customizing a runtime to use a MIG-enabled profile.

Use the following predefined hardware specifications for GPU inferencing:

Hardware specifications for GPU inferencing
Size Hardware definition
GPUx1 1GPU, 1 CPU and 4 GB RAM
GPUx2 2GPU, 2 CPU and 8 GB RAM
GPUx3 3GPU, 2 CPU and 12 GB RAM
GPUx4 4GPU, 2 CPU and 16 GB RAM

Learn more

Parent topic: Managing predictive deployments