Chapter 5. Software policies
5.1. Test Suite versions Copy linkLink copied to clipboard!
Red Hat recommends that the latest version of the test suite packages be used for all testing.
When a new version of any test suite package is made available, test results created using the previous versions will continue to be accepted for a period of three months. At the end of this period results from these previous versions will automatically be rejected and testing will need to be repeated with the later test suite packages.
The current valid package versions are displayed on the results package submission form.
5.2. Red Hat Enterprise Linux versions Copy linkLink copied to clipboard!
The latest minor release of Red Hat Enterprise Linux (RHEL) version is always recommended; however, any release that satisfies the full testing criteria may be used. Testing on the earliest fully-supported release will maximize the potential customer base. If multiple minor releases are used during testing, the newest minor release will be used as the posted release for the model. Depending on the features of a given model a minimum release may be required other than what is desired.
Instance Type that are available without certified images are required to be installed and tested using the GA media (or equivalent) without additional errata.
The test suite is tested against RHEL 8, RHEL 9, and RHEL 10 Base OS server installations. All the variants of the RHEL major version share a common core set of packages, and the use of these variants is allowed during certification testing. However, the variants may only provide a subset of the required packages that may result in the need for retesting with the Base OS variant.
Technical assistance during certification is not offered when using these variants.
5.3. Red Hat Enterprise Linux AI and Red Hat AI Inference Server Versions Copy linkLink copied to clipboard!
- Red Hat Enterprise Linux AI
- New Red Hat Enterprise Linux AI (RHEL AI) requests are accepted only on the current minor release of the RHEL AI product. When a new RHEL AI minor release becomes available, in progress certifications from the previous minor release must be completed within 30 days.
- Red Hat AI Inference Server
- New Red Hat AI Inference Server (RHAIIS) requests are accepted only on the current minor release of the RHAIIS product. When a new RHAIIS minor release becomes available, in progress certifications from the previous minor release must be completed within 30 days.
| Hardware Class | Catalog Features | Required Tests | Required Hardware |
|---|---|---|---|
| Data Center GPU [a] | AI Inferencing Accelerator | vllm_inferencing or Pytorch_inferencing | One member from each GPU design [b] configured with each logical compute units and sufficient memory [c] for inference testing. |
| AI Inference Accelerator [d] | AI Inferencing Accelerator | pytorch_inferencing | One member from each GPU design [e] configured with each logical compute units and sufficient memory [f] for inference testing. |
[a]
Available, but not required, in RHEL versions 9.0 and later certifications.
[b]
An accelerator family is considered to be any set of accelerators with the same design and generation, scaled up or down, that uses the same code path(s) in the same drivers and software stacks.
[c]
Different accelerator memory is not tracked in the context of instance certification.
[d]
Available, but not required, in RHEL versions 9.0 and later certifications.
[e]
An accelerator family is considered to be any set of accelerators with the same design and generation, scaled up or down, that uses the same code path(s) in the same drivers and software stacks.
[f]
Different accelerator memory is not tracked in the context of instance certification.
| |||
5.4. Unmodified Red Hat software Copy linkLink copied to clipboard!
To certify a Red Hat CCSP Instance Type, you must use a standard installation of Red Hat Enterprise Linux without making any modifications. You can adjust the default configuration set by the installer and first-boot utilities, only if you use standard system tools and the changes do not risk data loss. You must document any required changes in a Red Hat Knowledge Base Solution linked to your certification listing.
Do not modify the Red Hat certification test suite. The suite includes a self-check and will fail the supportable test if it detects any modifications.
If you are building bootc container images for RHEL AI certification, ensure the image includes only content from the corresponding Red Hat Enterprise Linux AI product version and associated downloads. Additional packages from other Red Hat products or modifications from outside of the base RHEL AI content are not allowed.
For cloud deployments, Red Hat provides official RHEL AI images on RHEL AI download page for AWS, Google Cloud, IBM Cloud, and Azure. Partners are encouraged to use these images when possible to streamline certification and ensure alignment with supported configurations.
For more detailed requirements and guidance on using bootc containers or cloud images with RHEL AI, refer to the RHEL AI installation documentation.
5.5. Kernel Boot parameters Copy linkLink copied to clipboard!
Additional kernel parameters may be utilized if they
- Are used to correct hardware configuration
- Do not disable functionality
- Do not expose a potential for data loss when not in use
Example
if the kernel parameter noacpi is required to boot a system which does not install without that parameter, this would likely be acceptable. If, however, the system would install but corrupts data over time when noacpi is not specified, the certification would be suspended until the situation is resolved. Additional kernel parameters utilized during certification are documented in Red Hat Knowledge Base Solution associated with the certification listing.
5.6. SELinux Copy linkLink copied to clipboard!
Certifications must be run with SELinux enabled using the Targeted Policy and with Enforcing on. The test suite will check for these conditions.