Worked on improving GPU checkpoint and restore reliability in the geerlingguy/linux repository, focusing on devices with multiple compute units. Addressed a bug affecting checkpoint-restore integrity by refining how queue checkpoint information is retrieved and ensuring all memory queue descriptors within a partition are correctly saved and restored. The solution involved targeted changes to the DRM AMD kernel driver, specifically within the amdkfd subsystem, to enhance resiliency during suspend and resume cycles. Utilized C for kernel development, applying expertise in device drivers and GPU programming to reduce checkpoint failures and maintain consistent state across GPU partitions in multi-xcc environments.