This document describes a requirement for the AMD Instinct™ Interlock-IOMMU passthrough setting for systems when Input-Output Memory Management Unit (IOMMU) is enabled, by default, in SBIOS.
Products Impacted
AMD Instinct™ and AMD Instinct accelerators, such as AMD Instinct™ using ROCm in a Linux environment.
Known Issue
Multi-GPU environments are failing due to limitations of the IOMMU enablement.
Recommended Solution
Disabling IOMMU using the grub command line resolves the issue.
Steps
To enable IOMMU
sudo bash -c 'echo GRUB_CMDLINE_LINUX="amd_iommu=on iommu=pt" >> /etc/default/grub'
sudo grub2-mkconfig -o /boot/efi/EFI/centos/grub.cfg
sudo reboot
To disable IOMMU
sudo bash -c 'echo GRUB_CMDLINE_LINUX="amd_iommu=off" >> /etc/default/grub'
sudo grub2-mkconfig -o /boot/efi/EFI/centos/grub.cfg
sudo reboot