CPU-only Deployment

Edit online

AI Services supports CPU-only deployments as an alternative to accelerator-based deployments. This enables users to run AI workloads on standard CPU hardware without requiring specialized accelerators such as IBM Spyre™.

By following the setup instructions, users can configure AI Services and launch the catalog. The catalog UI provides an interface to create, manage, and monitor applications within AI Services. CPU-only deployments are supported only with the Podman runtime. Containers deployed use default security contexts and do not require special SELinux policies or device access.

Use Cases for CPU-Only Deployments
  • Development & Testing: Quick setup for development environments.
  • Small-Scale Deployments: Suitable for low-volume production workloads.
  • Cost-Sensitive Scenarios: Ideal when specialized hardware is not available or not cost-effective.
  • Proof of Concept: Useful for evaluating AI Services before investing in accelerated infrastructure.
  • Edge Deployments: Enables running on standard server hardware without GPU requirements.
  • Fit-for-purpose deployments: Some AI Services run optimally as CPU-only deployment.