For deep learning tasks it would be great to have an easy way to use cuda devices in MS Fabric notebooks.
Processing Apache Spark SQL workloads on GPUs (ex: NVIDIA Rapids) is a way to return query results to customers much faster than CPUs.
Transformations routines (ELT or ETL) would speed up the time it takes to create datasets for customers too.
HPE Ezmeral Analytics examples (5 terabyte datasets) show on how it can be done on that platform.
CPU bound query engines like Apache Spark in Microsoft Fabric may not feed GPU, TPU, or NPU backed machine learning models fast enough.
