The Million Veteran Program (MVP) phenome-wide association study (PheWAS) represents a groundbreaking large-scale analysis of genome-wide genetic variation across a diverse catalog of phenotypes. It uniquely includes participants from populations historically underrepresented in genetic research, such as individuals with African and Admixed American genetic ancestry. This was made possible through the implementation of SAIGE, a state-of-the-art, multi-level modeling-based GWAS tool designed to account for sample relatedness and address the challenges of unbalanced case-control ratios, which are common in large-scale biobanks integrated with health systems. To enhance the efficiency of these analyses, we optimized the SAIGE algorithm to utilize GPU infrastructure, enabling faster matrix-vector operations. This was previously demonstrated on the OLCF Summit supercomputer using NVIDIA A100 GPU nodes. In the next phase of our project, supported by an INCITE award, we aim to further optimize SAIGE and expand the scale of our genomic association tests to include even more variants. These analyses will be conducted on the OLCF Frontier supercomputer, which features AMD GPU nodes similar to those available in the JLSE testbed. While our INCITE project officially begins in a few months, early access to the JLSE testbed offers a critical opportunity to adapt the SAIGE code—currently optimized for NVIDIA GPUs—to run effectively on AMD GPU nodes. This preparatory work will ensure a seamless transition and enhanced performance when the larger-scale analyses commence.