During testing of the Intel Gaudi image, it was observed that the partitioning layout follows the default RHEL installation scheme(default Anaconda installer scheme), which assigns 70GB to /dev/mapper/rhel-root. This may differ from the partitioning used in NVIDIA and AMD images, leading to potential inconsistencies:

• /dev/mapper/rhel-root was initially 70G, with 51G used, leaving only 20G free after downloading a 13GB model.
• This could lead to space constraints when working with large models.
• Other accelerator images (NVIDIA, AMD) may not have this issue, suggesting an inconsistency in partitioning across different RHEL AI images.

Expected Outcome:

• Ensure that disk partitioning is consistent across all RHEL AI accelerator images (NVIDIA, AMD, Intel).
• Verify that model storage locations have adequate space to avoid manual resizing.

cc rh-ee-aturate, brault@redhat.com