Gaudi reference installation link is just linked and no version description, which result in compatibility issues.
Considering using operator to provision and manage Gaudi drivers of cluster platform. Habana Gaudi firmware/driver are installed using script in each node. It’s tedious to login each node to install driver. And there are some compatibility issues between drivers and containers. Containers will crash with difference version for difference driver version.
Pods cannot be deployed with habana-container-runtime 1.5, but can be deployed with habana-container-runtime 1.6.
Gaudi reference installation link is just linked and no version description, which result in compatibility issues.
Considering using operator to provision and manage Gaudi drivers of cluster platform. Habana Gaudi firmware/driver are installed using script in each node. It’s tedious to login each node to install driver. And there are some compatibility issues between drivers and containers. Containers will crash with difference version for difference driver version.
Pods cannot be deployed with habana-container-runtime 1.5, but can be deployed with habana-container-runtime 1.6.