Researchers will use JLSE's H100 GPU testbed to run genome-scale variant effect prediction with Google DeepMind's AlphaGenome model across ~1 billion genetic variants. Most disease-associated variants fall in non-coding regulatory regions where their functional impact is poorly understood. AlphaGenome addresses this directly as it predicts regulatory and transcriptional effects of any DNA sequence variant with state-of-the-art accuracy across hundreds of cell types. The catch is cost: at population scale, inference is simply not feasible on CPU clusters or commodity GPUs. The H100 architecture is particularly well suited to this workload. The 80 GB HBM3 memory and 4th-gen tensor cores match AlphaGenome's memory footprint well, and NVLink enables the multi-GPU parallelism needed to process this dataset in weeks rather than years.