GustANN
GustANN is a high-throughput, cost-effective billion-scale vector search engine designed to run on a single GPU. Based on research published at SIGMOD 26, it leverages a graph-based architecture to achieve approximately 250,000 queries per second on billion-scale datasets like SIFT1B with 90 percent recall, outperforming existing solutions like DiskANN by a factor of 7.81. The system is engineered for memory efficiency, requiring only around 40GB of total memory for both GPU and CPU components even when handling vast datasets. It offers flexible deployment options, supporting search modes that are SSD-based, fully in DRAM, or entirely on the GPU, and integrates with various storage backends including SPDK, liburing, and libaio. Designed for x86 systems with NVIDIA GPUs such as the A100, GustANN supports datasets up to 2 billion vectors with record sizes under 4KB. It relies on DiskANN for index construction and features an automated setup process for rapid deployment. The software is suitable for large-scale