Running Low-Latency Analytical Workloads with GPU-Accelerated Presto on NVIDIA GB200 NVL72 - NVIDIA Developer

This article explores how GPU-accelerated Presto on NVIDIA GB200 NVL72 enables low-latency analytical workloads. Presto, an open-source distributed SQL engine, excels in processing large-scale dataset... Read original →

NVIDIA’s deep GPU integration of Presto marks the arrival of 'full-stack acceleration' in analytical databases. Technically, merging cuDF with Velox doesn’t just shrink ETL latency—it forces storage layers to adopt GPUDirect Storage, explaining IBM Storage Scale’s strategic inclusion as a signal of NVMe-over-Fabric displacing legacy HDFS. From a compliance angle, deploying such high-performance clusters in Taiwan, China or Southeast Asia may trigger new U.S. BIS export controls on advanced compute, compelling firms to redesign data-localization strategies. As Databricks and Snowflake double down on CPU-plus-vectorized hybrids, NVIDIA is staking claim to real-time decisioning—forcing rivals to either build native GPU schedulers or accept a widening performance gap. Within 18 months, open-source SQL engines will split: those lacking UCX support or NVLink topology awareness will fade, and TPC-H will shift from benchmark to baseline requirement.